├── codes.txt ├── README.md ├── valid.txt ├── LICENSE └── RoadSurfaceSegmentation.ipynb /codes.txt: -------------------------------------------------------------------------------- 1 | background 2 | roadAsphalt 3 | roadPaved 4 | roadUnpaved 5 | roadMarking 6 | speedBump 7 | catsEye 8 | stormDrain 9 | manholeCover 10 | patchs 11 | waterPuddle 12 | pothole 13 | craks 14 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # Road surface detection and differentiation considering surface damages 2 | 3 | [![Road Surface Semantic Segmentation](https://i.imgur.com/L0OFzOk.jpg)](https://youtu.be/uUchRHWOEF4 "Road Surface Semantic Segmentation") 4 | 5 | The semantic segmentation GT for road surfaces contains 701 frames from [RTK dataset](http://www.lapix.ufsc.br/pesquisas/projeto-veiculo-autonomo/datasets/?lang=en). Classes are defined as follows: 6 | 7 | - Background, everything being unrelated to the road surface; 8 | - Asphalt, roads with asphalt surface; 9 | - Paved, different pavements (eg.: Cobblestone); 10 | - Unpaved, for unpaved roads; 11 | - Markings, to the road markings; 12 | - Speed-Bump, for the speed-bumps on the road; 13 | - Cats-Eye, for the cats-eye found on the road, both on the side and in the center of the path; 14 | - Storm-Drain, usually at the side edges of the road; 15 | - Patch, for the various patches found on asphalt road; 16 | - Water-Puddle, we use this class also for muddy regions; 17 | - Pothole, for different types and sizes of potholes, no matter if they are on asphalt, paved or unpaved roads; 18 | - Cracks, used in different road damages, like ruptures. 19 | 20 | ## Citation: 21 | ``` 22 | @misc{rateke:2020.3, 23 | title = {Road surface detection and differentiation considering surface damages}, 24 | author = {Thiago Rateke and Aldo von Wangenheim}, 25 | journal={Autonomous Robots}, 26 | year={2021}, 27 | month={Jan}, 28 | day={11}, 29 | issn={1573-7527}, 30 | doi={10.1007/s10514-020-09964-3}, 31 | url={https://doi.org/10.1007/s10514-020-09964-3} 32 | } 33 | ``` 34 | - [Original Paper](https://rdcu.be/cdpxi) 35 | - [Step-by-step](https://towardsdatascience.com/road-surface-semantic-segmentation-4d65b045245?source=friends_link&sk=f1fbe72fecd65331eb326a513b1f464f) 36 | -------------------------------------------------------------------------------- /valid.txt: -------------------------------------------------------------------------------- 1 | 000000001.png 2 | 000000004.png 3 | 000000010.png 4 | 000000012.png 5 | 000000019.png 6 | 000000025.png 7 | 000000029.png 8 | 000000038.png 9 | 000000045.png 10 | 000000050.png 11 | 000000052.png 12 | 000000060.png 13 | 000000072.png 14 | 000000079.png 15 | 000000085.png 16 | 000000087.png 17 | 000000094.png 18 | 000000100.png 19 | 000000103.png 20 | 000000106.png 21 | 000000116.png 22 | 000000120.png 23 | 000000123.png 24 | 000000130.png 25 | 000000134.png 26 | 000000137.png 27 | 000000150.png 28 | 000000152.png 29 | 000000159.png 30 | 000000178.png 31 | 000000180.png 32 | 000000186.png 33 | 000000192.png 34 | 000000194.png 35 | 000000198.png 36 | 000000204.png 37 | 000000207.png 38 | 000000214.png 39 | 000000221.png 40 | 000000224.png 41 | 000000225.png 42 | 000000228.png 43 | 000000231.png 44 | 000000233.png 45 | 000000237.png 46 | 000000240.png 47 | 000000243.png 48 | 000000244.png 49 | 000000248.png 50 | 000000250.png 51 | 000000265.png 52 | 000000266.png 53 | 000000269.png 54 | 000000272.png 55 | 000000276.png 56 | 000000282.png 57 | 000000284.png 58 | 000000289.png 59 | 000000290.png 60 | 000000298.png 61 | 000000299.png 62 | 000000323.png 63 | 000000327.png 64 | 000000339.png 65 | 000000344.png 66 | 000000346.png 67 | 000000351.png 68 | 000000356.png 69 | 000000365.png 70 | 000000367.png 71 | 000000374.png 72 | 000000381.png 73 | 000000395.png 74 | 000000398.png 75 | 000000399.png 76 | 000000402.png 77 | 000000407.png 78 | 000000409.png 79 | 000000412.png 80 | 000000414.png 81 | 000000417.png 82 | 000000423.png 83 | 000000425.png 84 | 000000429.png 85 | 000000430.png 86 | 000000439.png 87 | 000000447.png 88 | 000000449.png 89 | 000000451.png 90 | 000000460.png 91 | 000000463.png 92 | 000000465.png 93 | 000000471.png 94 | 000000472.png 95 | 000000479.png 96 | 000000481.png 97 | 000000486.png 98 | 000000500.png 99 | 000000502.png 100 | 000000506.png 101 | 000000507.png 102 | 000000515.png 103 | 000000517.png 104 | 000000522.png 105 | 000000527.png 106 | 000000529.png 107 | 000000543.png 108 | 000000549.png 109 | 000000556.png 110 | 000000563.png 111 | 000000577.png 112 | 000000598.png 113 | 000000602.png 114 | 000000605.png 115 | 000000612.png 116 | 000000620.png 117 | 000000626.png 118 | 000000632.png 119 | 000000635.png 120 | 000000642.png 121 | 000000646.png 122 | 000000653.png 123 | 000000655.png 124 | 000000656.png 125 | 000000660.png 126 | 000000662.png 127 | 000000663.png 128 | 000000666.png 129 | 000000668.png 130 | 000000669.png 131 | 000000675.png 132 | 000000678.png 133 | 000000681.png 134 | 000000684.png 135 | 000000685.png 136 | 000000687.png 137 | 000000690.png 138 | 000000694.png 139 | 000000698.png 140 | 000000699.png 141 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | GNU GENERAL PUBLIC LICENSE 2 | Version 3, 29 June 2007 3 | 4 | Copyright (C) 2007 Free Software Foundation, Inc. 5 | Everyone is permitted to copy and distribute verbatim copies 6 | of this license document, but changing it is not allowed. 7 | 8 | Preamble 9 | 10 | The GNU General Public License is a free, copyleft license for 11 | software and other kinds of works. 12 | 13 | The licenses for most software and other practical works are designed 14 | to take away your freedom to share and change the works. By contrast, 15 | the GNU General Public License is intended to guarantee your freedom to 16 | share and change all versions of a program--to make sure it remains free 17 | software for all its users. We, the Free Software Foundation, use the 18 | GNU General Public License for most of our software; it applies also to 19 | any other work released this way by its authors. You can apply it to 20 | your programs, too. 21 | 22 | When we speak of free software, we are referring to freedom, not 23 | price. Our General Public Licenses are designed to make sure that you 24 | have the freedom to distribute copies of free software (and charge for 25 | them if you wish), that you receive source code or can get it if you 26 | want it, that you can change the software or use pieces of it in new 27 | free programs, and that you know you can do these things. 28 | 29 | To protect your rights, we need to prevent others from denying you 30 | these rights or asking you to surrender the rights. Therefore, you have 31 | certain responsibilities if you distribute copies of the software, or if 32 | you modify it: responsibilities to respect the freedom of others. 33 | 34 | For example, if you distribute copies of such a program, whether 35 | gratis or for a fee, you must pass on to the recipients the same 36 | freedoms that you received. You must make sure that they, too, receive 37 | or can get the source code. And you must show them these terms so they 38 | know their rights. 39 | 40 | Developers that use the GNU GPL protect your rights with two steps: 41 | (1) assert copyright on the software, and (2) offer you this License 42 | giving you legal permission to copy, distribute and/or modify it. 43 | 44 | For the developers' and authors' protection, the GPL clearly explains 45 | that there is no warranty for this free software. For both users' and 46 | authors' sake, the GPL requires that modified versions be marked as 47 | changed, so that their problems will not be attributed erroneously to 48 | authors of previous versions. 49 | 50 | Some devices are designed to deny users access to install or run 51 | modified versions of the software inside them, although the manufacturer 52 | can do so. This is fundamentally incompatible with the aim of 53 | protecting users' freedom to change the software. The systematic 54 | pattern of such abuse occurs in the area of products for individuals to 55 | use, which is precisely where it is most unacceptable. Therefore, we 56 | have designed this version of the GPL to prohibit the practice for those 57 | products. If such problems arise substantially in other domains, we 58 | stand ready to extend this provision to those domains in future versions 59 | of the GPL, as needed to protect the freedom of users. 60 | 61 | Finally, every program is threatened constantly by software patents. 62 | States should not allow patents to restrict development and use of 63 | software on general-purpose computers, but in those that do, we wish to 64 | avoid the special danger that patents applied to a free program could 65 | make it effectively proprietary. To prevent this, the GPL assures that 66 | patents cannot be used to render the program non-free. 67 | 68 | The precise terms and conditions for copying, distribution and 69 | modification follow. 70 | 71 | TERMS AND CONDITIONS 72 | 73 | 0. Definitions. 74 | 75 | "This License" refers to version 3 of the GNU General Public License. 76 | 77 | "Copyright" also means copyright-like laws that apply to other kinds of 78 | works, such as semiconductor masks. 79 | 80 | "The Program" refers to any copyrightable work licensed under this 81 | License. Each licensee is addressed as "you". "Licensees" and 82 | "recipients" may be individuals or organizations. 83 | 84 | To "modify" a work means to copy from or adapt all or part of the work 85 | in a fashion requiring copyright permission, other than the making of an 86 | exact copy. The resulting work is called a "modified version" of the 87 | earlier work or a work "based on" the earlier work. 88 | 89 | A "covered work" means either the unmodified Program or a work based 90 | on the Program. 91 | 92 | To "propagate" a work means to do anything with it that, without 93 | permission, would make you directly or secondarily liable for 94 | infringement under applicable copyright law, except executing it on a 95 | computer or modifying a private copy. Propagation includes copying, 96 | distribution (with or without modification), making available to the 97 | public, and in some countries other activities as well. 98 | 99 | To "convey" a work means any kind of propagation that enables other 100 | parties to make or receive copies. Mere interaction with a user through 101 | a computer network, with no transfer of a copy, is not conveying. 102 | 103 | An interactive user interface displays "Appropriate Legal Notices" 104 | to the extent that it includes a convenient and prominently visible 105 | feature that (1) displays an appropriate copyright notice, and (2) 106 | tells the user that there is no warranty for the work (except to the 107 | extent that warranties are provided), that licensees may convey the 108 | work under this License, and how to view a copy of this License. If 109 | the interface presents a list of user commands or options, such as a 110 | menu, a prominent item in the list meets this criterion. 111 | 112 | 1. Source Code. 113 | 114 | The "source code" for a work means the preferred form of the work 115 | for making modifications to it. "Object code" means any non-source 116 | form of a work. 117 | 118 | A "Standard Interface" means an interface that either is an official 119 | standard defined by a recognized standards body, or, in the case of 120 | interfaces specified for a particular programming language, one that 121 | is widely used among developers working in that language. 122 | 123 | The "System Libraries" of an executable work include anything, other 124 | than the work as a whole, that (a) is included in the normal form of 125 | packaging a Major Component, but which is not part of that Major 126 | Component, and (b) serves only to enable use of the work with that 127 | Major Component, or to implement a Standard Interface for which an 128 | implementation is available to the public in source code form. A 129 | "Major Component", in this context, means a major essential component 130 | (kernel, window system, and so on) of the specific operating system 131 | (if any) on which the executable work runs, or a compiler used to 132 | produce the work, or an object code interpreter used to run it. 133 | 134 | The "Corresponding Source" for a work in object code form means all 135 | the source code needed to generate, install, and (for an executable 136 | work) run the object code and to modify the work, including scripts to 137 | control those activities. However, it does not include the work's 138 | System Libraries, or general-purpose tools or generally available free 139 | programs which are used unmodified in performing those activities but 140 | which are not part of the work. For example, Corresponding Source 141 | includes interface definition files associated with source files for 142 | the work, and the source code for shared libraries and dynamically 143 | linked subprograms that the work is specifically designed to require, 144 | such as by intimate data communication or control flow between those 145 | subprograms and other parts of the work. 146 | 147 | The Corresponding Source need not include anything that users 148 | can regenerate automatically from other parts of the Corresponding 149 | Source. 150 | 151 | The Corresponding Source for a work in source code form is that 152 | same work. 153 | 154 | 2. Basic Permissions. 155 | 156 | All rights granted under this License are granted for the term of 157 | copyright on the Program, and are irrevocable provided the stated 158 | conditions are met. This License explicitly affirms your unlimited 159 | permission to run the unmodified Program. The output from running a 160 | covered work is covered by this License only if the output, given its 161 | content, constitutes a covered work. This License acknowledges your 162 | rights of fair use or other equivalent, as provided by copyright law. 163 | 164 | You may make, run and propagate covered works that you do not 165 | convey, without conditions so long as your license otherwise remains 166 | in force. You may convey covered works to others for the sole purpose 167 | of having them make modifications exclusively for you, or provide you 168 | with facilities for running those works, provided that you comply with 169 | the terms of this License in conveying all material for which you do 170 | not control copyright. Those thus making or running the covered works 171 | for you must do so exclusively on your behalf, under your direction 172 | and control, on terms that prohibit them from making any copies of 173 | your copyrighted material outside their relationship with you. 174 | 175 | Conveying under any other circumstances is permitted solely under 176 | the conditions stated below. Sublicensing is not allowed; section 10 177 | makes it unnecessary. 178 | 179 | 3. Protecting Users' Legal Rights From Anti-Circumvention Law. 180 | 181 | No covered work shall be deemed part of an effective technological 182 | measure under any applicable law fulfilling obligations under article 183 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or 184 | similar laws prohibiting or restricting circumvention of such 185 | measures. 186 | 187 | When you convey a covered work, you waive any legal power to forbid 188 | circumvention of technological measures to the extent such circumvention 189 | is effected by exercising rights under this License with respect to 190 | the covered work, and you disclaim any intention to limit operation or 191 | modification of the work as a means of enforcing, against the work's 192 | users, your or third parties' legal rights to forbid circumvention of 193 | technological measures. 194 | 195 | 4. Conveying Verbatim Copies. 196 | 197 | You may convey verbatim copies of the Program's source code as you 198 | receive it, in any medium, provided that you conspicuously and 199 | appropriately publish on each copy an appropriate copyright notice; 200 | keep intact all notices stating that this License and any 201 | non-permissive terms added in accord with section 7 apply to the code; 202 | keep intact all notices of the absence of any warranty; and give all 203 | recipients a copy of this License along with the Program. 204 | 205 | You may charge any price or no price for each copy that you convey, 206 | and you may offer support or warranty protection for a fee. 207 | 208 | 5. Conveying Modified Source Versions. 209 | 210 | You may convey a work based on the Program, or the modifications to 211 | produce it from the Program, in the form of source code under the 212 | terms of section 4, provided that you also meet all of these conditions: 213 | 214 | a) The work must carry prominent notices stating that you modified 215 | it, and giving a relevant date. 216 | 217 | b) The work must carry prominent notices stating that it is 218 | released under this License and any conditions added under section 219 | 7. This requirement modifies the requirement in section 4 to 220 | "keep intact all notices". 221 | 222 | c) You must license the entire work, as a whole, under this 223 | License to anyone who comes into possession of a copy. This 224 | License will therefore apply, along with any applicable section 7 225 | additional terms, to the whole of the work, and all its parts, 226 | regardless of how they are packaged. This License gives no 227 | permission to license the work in any other way, but it does not 228 | invalidate such permission if you have separately received it. 229 | 230 | d) If the work has interactive user interfaces, each must display 231 | Appropriate Legal Notices; however, if the Program has interactive 232 | interfaces that do not display Appropriate Legal Notices, your 233 | work need not make them do so. 234 | 235 | A compilation of a covered work with other separate and independent 236 | works, which are not by their nature extensions of the covered work, 237 | and which are not combined with it such as to form a larger program, 238 | in or on a volume of a storage or distribution medium, is called an 239 | "aggregate" if the compilation and its resulting copyright are not 240 | used to limit the access or legal rights of the compilation's users 241 | beyond what the individual works permit. Inclusion of a covered work 242 | in an aggregate does not cause this License to apply to the other 243 | parts of the aggregate. 244 | 245 | 6. Conveying Non-Source Forms. 246 | 247 | You may convey a covered work in object code form under the terms 248 | of sections 4 and 5, provided that you also convey the 249 | machine-readable Corresponding Source under the terms of this License, 250 | in one of these ways: 251 | 252 | a) Convey the object code in, or embodied in, a physical product 253 | (including a physical distribution medium), accompanied by the 254 | Corresponding Source fixed on a durable physical medium 255 | customarily used for software interchange. 256 | 257 | b) Convey the object code in, or embodied in, a physical product 258 | (including a physical distribution medium), accompanied by a 259 | written offer, valid for at least three years and valid for as 260 | long as you offer spare parts or customer support for that product 261 | model, to give anyone who possesses the object code either (1) a 262 | copy of the Corresponding Source for all the software in the 263 | product that is covered by this License, on a durable physical 264 | medium customarily used for software interchange, for a price no 265 | more than your reasonable cost of physically performing this 266 | conveying of source, or (2) access to copy the 267 | Corresponding Source from a network server at no charge. 268 | 269 | c) Convey individual copies of the object code with a copy of the 270 | written offer to provide the Corresponding Source. This 271 | alternative is allowed only occasionally and noncommercially, and 272 | only if you received the object code with such an offer, in accord 273 | with subsection 6b. 274 | 275 | d) Convey the object code by offering access from a designated 276 | place (gratis or for a charge), and offer equivalent access to the 277 | Corresponding Source in the same way through the same place at no 278 | further charge. You need not require recipients to copy the 279 | Corresponding Source along with the object code. If the place to 280 | copy the object code is a network server, the Corresponding Source 281 | may be on a different server (operated by you or a third party) 282 | that supports equivalent copying facilities, provided you maintain 283 | clear directions next to the object code saying where to find the 284 | Corresponding Source. Regardless of what server hosts the 285 | Corresponding Source, you remain obligated to ensure that it is 286 | available for as long as needed to satisfy these requirements. 287 | 288 | e) Convey the object code using peer-to-peer transmission, provided 289 | you inform other peers where the object code and Corresponding 290 | Source of the work are being offered to the general public at no 291 | charge under subsection 6d. 292 | 293 | A separable portion of the object code, whose source code is excluded 294 | from the Corresponding Source as a System Library, need not be 295 | included in conveying the object code work. 296 | 297 | A "User Product" is either (1) a "consumer product", which means any 298 | tangible personal property which is normally used for personal, family, 299 | or household purposes, or (2) anything designed or sold for incorporation 300 | into a dwelling. In determining whether a product is a consumer product, 301 | doubtful cases shall be resolved in favor of coverage. For a particular 302 | product received by a particular user, "normally used" refers to a 303 | typical or common use of that class of product, regardless of the status 304 | of the particular user or of the way in which the particular user 305 | actually uses, or expects or is expected to use, the product. A product 306 | is a consumer product regardless of whether the product has substantial 307 | commercial, industrial or non-consumer uses, unless such uses represent 308 | the only significant mode of use of the product. 309 | 310 | "Installation Information" for a User Product means any methods, 311 | procedures, authorization keys, or other information required to install 312 | and execute modified versions of a covered work in that User Product from 313 | a modified version of its Corresponding Source. The information must 314 | suffice to ensure that the continued functioning of the modified object 315 | code is in no case prevented or interfered with solely because 316 | modification has been made. 317 | 318 | If you convey an object code work under this section in, or with, or 319 | specifically for use in, a User Product, and the conveying occurs as 320 | part of a transaction in which the right of possession and use of the 321 | User Product is transferred to the recipient in perpetuity or for a 322 | fixed term (regardless of how the transaction is characterized), the 323 | Corresponding Source conveyed under this section must be accompanied 324 | by the Installation Information. But this requirement does not apply 325 | if neither you nor any third party retains the ability to install 326 | modified object code on the User Product (for example, the work has 327 | been installed in ROM). 328 | 329 | The requirement to provide Installation Information does not include a 330 | requirement to continue to provide support service, warranty, or updates 331 | for a work that has been modified or installed by the recipient, or for 332 | the User Product in which it has been modified or installed. Access to a 333 | network may be denied when the modification itself materially and 334 | adversely affects the operation of the network or violates the rules and 335 | protocols for communication across the network. 336 | 337 | Corresponding Source conveyed, and Installation Information provided, 338 | in accord with this section must be in a format that is publicly 339 | documented (and with an implementation available to the public in 340 | source code form), and must require no special password or key for 341 | unpacking, reading or copying. 342 | 343 | 7. Additional Terms. 344 | 345 | "Additional permissions" are terms that supplement the terms of this 346 | License by making exceptions from one or more of its conditions. 347 | Additional permissions that are applicable to the entire Program shall 348 | be treated as though they were included in this License, to the extent 349 | that they are valid under applicable law. If additional permissions 350 | apply only to part of the Program, that part may be used separately 351 | under those permissions, but the entire Program remains governed by 352 | this License without regard to the additional permissions. 353 | 354 | When you convey a copy of a covered work, you may at your option 355 | remove any additional permissions from that copy, or from any part of 356 | it. (Additional permissions may be written to require their own 357 | removal in certain cases when you modify the work.) You may place 358 | additional permissions on material, added by you to a covered work, 359 | for which you have or can give appropriate copyright permission. 360 | 361 | Notwithstanding any other provision of this License, for material you 362 | add to a covered work, you may (if authorized by the copyright holders of 363 | that material) supplement the terms of this License with terms: 364 | 365 | a) Disclaiming warranty or limiting liability differently from the 366 | terms of sections 15 and 16 of this License; or 367 | 368 | b) Requiring preservation of specified reasonable legal notices or 369 | author attributions in that material or in the Appropriate Legal 370 | Notices displayed by works containing it; or 371 | 372 | c) Prohibiting misrepresentation of the origin of that material, or 373 | requiring that modified versions of such material be marked in 374 | reasonable ways as different from the original version; or 375 | 376 | d) Limiting the use for publicity purposes of names of licensors or 377 | authors of the material; or 378 | 379 | e) Declining to grant rights under trademark law for use of some 380 | trade names, trademarks, or service marks; or 381 | 382 | f) Requiring indemnification of licensors and authors of that 383 | material by anyone who conveys the material (or modified versions of 384 | it) with contractual assumptions of liability to the recipient, for 385 | any liability that these contractual assumptions directly impose on 386 | those licensors and authors. 387 | 388 | All other non-permissive additional terms are considered "further 389 | restrictions" within the meaning of section 10. If the Program as you 390 | received it, or any part of it, contains a notice stating that it is 391 | governed by this License along with a term that is a further 392 | restriction, you may remove that term. If a license document contains 393 | a further restriction but permits relicensing or conveying under this 394 | License, you may add to a covered work material governed by the terms 395 | of that license document, provided that the further restriction does 396 | not survive such relicensing or conveying. 397 | 398 | If you add terms to a covered work in accord with this section, you 399 | must place, in the relevant source files, a statement of the 400 | additional terms that apply to those files, or a notice indicating 401 | where to find the applicable terms. 402 | 403 | Additional terms, permissive or non-permissive, may be stated in the 404 | form of a separately written license, or stated as exceptions; 405 | the above requirements apply either way. 406 | 407 | 8. Termination. 408 | 409 | You may not propagate or modify a covered work except as expressly 410 | provided under this License. Any attempt otherwise to propagate or 411 | modify it is void, and will automatically terminate your rights under 412 | this License (including any patent licenses granted under the third 413 | paragraph of section 11). 414 | 415 | However, if you cease all violation of this License, then your 416 | license from a particular copyright holder is reinstated (a) 417 | provisionally, unless and until the copyright holder explicitly and 418 | finally terminates your license, and (b) permanently, if the copyright 419 | holder fails to notify you of the violation by some reasonable means 420 | prior to 60 days after the cessation. 421 | 422 | Moreover, your license from a particular copyright holder is 423 | reinstated permanently if the copyright holder notifies you of the 424 | violation by some reasonable means, this is the first time you have 425 | received notice of violation of this License (for any work) from that 426 | copyright holder, and you cure the violation prior to 30 days after 427 | your receipt of the notice. 428 | 429 | Termination of your rights under this section does not terminate the 430 | licenses of parties who have received copies or rights from you under 431 | this License. If your rights have been terminated and not permanently 432 | reinstated, you do not qualify to receive new licenses for the same 433 | material under section 10. 434 | 435 | 9. Acceptance Not Required for Having Copies. 436 | 437 | You are not required to accept this License in order to receive or 438 | run a copy of the Program. Ancillary propagation of a covered work 439 | occurring solely as a consequence of using peer-to-peer transmission 440 | to receive a copy likewise does not require acceptance. However, 441 | nothing other than this License grants you permission to propagate or 442 | modify any covered work. These actions infringe copyright if you do 443 | not accept this License. Therefore, by modifying or propagating a 444 | covered work, you indicate your acceptance of this License to do so. 445 | 446 | 10. Automatic Licensing of Downstream Recipients. 447 | 448 | Each time you convey a covered work, the recipient automatically 449 | receives a license from the original licensors, to run, modify and 450 | propagate that work, subject to this License. You are not responsible 451 | for enforcing compliance by third parties with this License. 452 | 453 | An "entity transaction" is a transaction transferring control of an 454 | organization, or substantially all assets of one, or subdividing an 455 | organization, or merging organizations. If propagation of a covered 456 | work results from an entity transaction, each party to that 457 | transaction who receives a copy of the work also receives whatever 458 | licenses to the work the party's predecessor in interest had or could 459 | give under the previous paragraph, plus a right to possession of the 460 | Corresponding Source of the work from the predecessor in interest, if 461 | the predecessor has it or can get it with reasonable efforts. 462 | 463 | You may not impose any further restrictions on the exercise of the 464 | rights granted or affirmed under this License. For example, you may 465 | not impose a license fee, royalty, or other charge for exercise of 466 | rights granted under this License, and you may not initiate litigation 467 | (including a cross-claim or counterclaim in a lawsuit) alleging that 468 | any patent claim is infringed by making, using, selling, offering for 469 | sale, or importing the Program or any portion of it. 470 | 471 | 11. Patents. 472 | 473 | A "contributor" is a copyright holder who authorizes use under this 474 | License of the Program or a work on which the Program is based. The 475 | work thus licensed is called the contributor's "contributor version". 476 | 477 | A contributor's "essential patent claims" are all patent claims 478 | owned or controlled by the contributor, whether already acquired or 479 | hereafter acquired, that would be infringed by some manner, permitted 480 | by this License, of making, using, or selling its contributor version, 481 | but do not include claims that would be infringed only as a 482 | consequence of further modification of the contributor version. For 483 | purposes of this definition, "control" includes the right to grant 484 | patent sublicenses in a manner consistent with the requirements of 485 | this License. 486 | 487 | Each contributor grants you a non-exclusive, worldwide, royalty-free 488 | patent license under the contributor's essential patent claims, to 489 | make, use, sell, offer for sale, import and otherwise run, modify and 490 | propagate the contents of its contributor version. 491 | 492 | In the following three paragraphs, a "patent license" is any express 493 | agreement or commitment, however denominated, not to enforce a patent 494 | (such as an express permission to practice a patent or covenant not to 495 | sue for patent infringement). To "grant" such a patent license to a 496 | party means to make such an agreement or commitment not to enforce a 497 | patent against the party. 498 | 499 | If you convey a covered work, knowingly relying on a patent license, 500 | and the Corresponding Source of the work is not available for anyone 501 | to copy, free of charge and under the terms of this License, through a 502 | publicly available network server or other readily accessible means, 503 | then you must either (1) cause the Corresponding Source to be so 504 | available, or (2) arrange to deprive yourself of the benefit of the 505 | patent license for this particular work, or (3) arrange, in a manner 506 | consistent with the requirements of this License, to extend the patent 507 | license to downstream recipients. "Knowingly relying" means you have 508 | actual knowledge that, but for the patent license, your conveying the 509 | covered work in a country, or your recipient's use of the covered work 510 | in a country, would infringe one or more identifiable patents in that 511 | country that you have reason to believe are valid. 512 | 513 | If, pursuant to or in connection with a single transaction or 514 | arrangement, you convey, or propagate by procuring conveyance of, a 515 | covered work, and grant a patent license to some of the parties 516 | receiving the covered work authorizing them to use, propagate, modify 517 | or convey a specific copy of the covered work, then the patent license 518 | you grant is automatically extended to all recipients of the covered 519 | work and works based on it. 520 | 521 | A patent license is "discriminatory" if it does not include within 522 | the scope of its coverage, prohibits the exercise of, or is 523 | conditioned on the non-exercise of one or more of the rights that are 524 | specifically granted under this License. You may not convey a covered 525 | work if you are a party to an arrangement with a third party that is 526 | in the business of distributing software, under which you make payment 527 | to the third party based on the extent of your activity of conveying 528 | the work, and under which the third party grants, to any of the 529 | parties who would receive the covered work from you, a discriminatory 530 | patent license (a) in connection with copies of the covered work 531 | conveyed by you (or copies made from those copies), or (b) primarily 532 | for and in connection with specific products or compilations that 533 | contain the covered work, unless you entered into that arrangement, 534 | or that patent license was granted, prior to 28 March 2007. 535 | 536 | Nothing in this License shall be construed as excluding or limiting 537 | any implied license or other defenses to infringement that may 538 | otherwise be available to you under applicable patent law. 539 | 540 | 12. No Surrender of Others' Freedom. 541 | 542 | If conditions are imposed on you (whether by court order, agreement or 543 | otherwise) that contradict the conditions of this License, they do not 544 | excuse you from the conditions of this License. If you cannot convey a 545 | covered work so as to satisfy simultaneously your obligations under this 546 | License and any other pertinent obligations, then as a consequence you may 547 | not convey it at all. For example, if you agree to terms that obligate you 548 | to collect a royalty for further conveying from those to whom you convey 549 | the Program, the only way you could satisfy both those terms and this 550 | License would be to refrain entirely from conveying the Program. 551 | 552 | 13. Use with the GNU Affero General Public License. 553 | 554 | Notwithstanding any other provision of this License, you have 555 | permission to link or combine any covered work with a work licensed 556 | under version 3 of the GNU Affero General Public License into a single 557 | combined work, and to convey the resulting work. The terms of this 558 | License will continue to apply to the part which is the covered work, 559 | but the special requirements of the GNU Affero General Public License, 560 | section 13, concerning interaction through a network will apply to the 561 | combination as such. 562 | 563 | 14. Revised Versions of this License. 564 | 565 | The Free Software Foundation may publish revised and/or new versions of 566 | the GNU General Public License from time to time. Such new versions will 567 | be similar in spirit to the present version, but may differ in detail to 568 | address new problems or concerns. 569 | 570 | Each version is given a distinguishing version number. If the 571 | Program specifies that a certain numbered version of the GNU General 572 | Public License "or any later version" applies to it, you have the 573 | option of following the terms and conditions either of that numbered 574 | version or of any later version published by the Free Software 575 | Foundation. If the Program does not specify a version number of the 576 | GNU General Public License, you may choose any version ever published 577 | by the Free Software Foundation. 578 | 579 | If the Program specifies that a proxy can decide which future 580 | versions of the GNU General Public License can be used, that proxy's 581 | public statement of acceptance of a version permanently authorizes you 582 | to choose that version for the Program. 583 | 584 | Later license versions may give you additional or different 585 | permissions. However, no additional obligations are imposed on any 586 | author or copyright holder as a result of your choosing to follow a 587 | later version. 588 | 589 | 15. Disclaimer of Warranty. 590 | 591 | THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY 592 | APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT 593 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY 594 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, 595 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 596 | PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM 597 | IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF 598 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION. 599 | 600 | 16. Limitation of Liability. 601 | 602 | IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING 603 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS 604 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY 605 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE 606 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF 607 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD 608 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), 609 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF 610 | SUCH DAMAGES. 611 | 612 | 17. Interpretation of Sections 15 and 16. 613 | 614 | If the disclaimer of warranty and limitation of liability provided 615 | above cannot be given local legal effect according to their terms, 616 | reviewing courts shall apply local law that most closely approximates 617 | an absolute waiver of all civil liability in connection with the 618 | Program, unless a warranty or assumption of liability accompanies a 619 | copy of the Program in return for a fee. 620 | 621 | END OF TERMS AND CONDITIONS 622 | 623 | How to Apply These Terms to Your New Programs 624 | 625 | If you develop a new program, and you want it to be of the greatest 626 | possible use to the public, the best way to achieve this is to make it 627 | free software which everyone can redistribute and change under these terms. 628 | 629 | To do so, attach the following notices to the program. It is safest 630 | to attach them to the start of each source file to most effectively 631 | state the exclusion of warranty; and each file should have at least 632 | the "copyright" line and a pointer to where the full notice is found. 633 | 634 | 635 | Copyright (C) 636 | 637 | This program is free software: you can redistribute it and/or modify 638 | it under the terms of the GNU General Public License as published by 639 | the Free Software Foundation, either version 3 of the License, or 640 | (at your option) any later version. 641 | 642 | This program is distributed in the hope that it will be useful, 643 | but WITHOUT ANY WARRANTY; without even the implied warranty of 644 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 645 | GNU General Public License for more details. 646 | 647 | You should have received a copy of the GNU General Public License 648 | along with this program. If not, see . 649 | 650 | Also add information on how to contact you by electronic and paper mail. 651 | 652 | If the program does terminal interaction, make it output a short 653 | notice like this when it starts in an interactive mode: 654 | 655 | Copyright (C) 656 | This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'. 657 | This is free software, and you are welcome to redistribute it 658 | under certain conditions; type `show c' for details. 659 | 660 | The hypothetical commands `show w' and `show c' should show the appropriate 661 | parts of the General Public License. Of course, your program's commands 662 | might be different; for a GUI interface, you would use an "about box". 663 | 664 | You should also get your employer (if you work as a programmer) or school, 665 | if any, to sign a "copyright disclaimer" for the program, if necessary. 666 | For more information on this, and how to apply and follow the GNU GPL, see 667 | . 668 | 669 | The GNU General Public License does not permit incorporating your program 670 | into proprietary programs. If your program is a subroutine library, you 671 | may consider it more useful to permit linking proprietary applications with 672 | the library. If this is what you want to do, use the GNU Lesser General 673 | Public License instead of this License. But first, please read 674 | . 675 | -------------------------------------------------------------------------------- /RoadSurfaceSegmentation.ipynb: -------------------------------------------------------------------------------- 1 | { 2 | "nbformat": 4, 3 | "nbformat_minor": 0, 4 | "metadata": { 5 | "colab": { 6 | "name": "Copy of RoadSurfaceSegmentation.ipynb", 7 | "provenance": [], 8 | "collapsed_sections": [ 9 | "mk2r-gCRCVb_", 10 | "nBWj9_LSC2G5", 11 | "UbxyIcZkDqBJ", 12 | "K7W20An-GnNm", 13 | "VE4zF162G-4z", 14 | "tRZoEiB8HJhI", 15 | "3oW77qUxHhkp", 16 | "3ojFnvRBHx8v", 17 | "KgZkM8M1IAxD", 18 | "_siWJq89IOdJ", 19 | "Uh1ZPx64gR56", 20 | "aT8l81QTjBrE", 21 | "33FD1Sc2j-k0", 22 | "4TAIZ9s4msq5", 23 | "D8N5E7BxnK7n", 24 | "XI-bgt6voXi9", 25 | "wAZbjvb40iMJ", 26 | "g5RVRyXL0_Uh", 27 | "eZ7SZZBN1KrS", 28 | "S0FHwE1p1RmR", 29 | "IO3-KXDsp4VB", 30 | "Ke8FATsZW1n_" 31 | ] 32 | }, 33 | "kernelspec": { 34 | "name": "python3", 35 | "display_name": "Python 3" 36 | }, 37 | "accelerator": "GPU" 38 | }, 39 | "cells": [ 40 | { 41 | "cell_type": "markdown", 42 | "metadata": { 43 | "id": "GGg9kEr0mzrf", 44 | "colab_type": "text" 45 | }, 46 | "source": [ 47 | "# Road surface detection and differentiation considering surface damages" 48 | ] 49 | }, 50 | { 51 | "cell_type": "markdown", 52 | "metadata": { 53 | "id": "mk2r-gCRCVb_", 54 | "colab_type": "text" 55 | }, 56 | "source": [ 57 | "## 1. Initial settings\n", 58 | "\n", 59 | "---\n", 60 | "\n" 61 | ] 62 | }, 63 | { 64 | "cell_type": "code", 65 | "metadata": { 66 | "id": "woWwcX5eznYL", 67 | "colab_type": "code", 68 | "colab": {} 69 | }, 70 | "source": [ 71 | "%reload_ext autoreload\n", 72 | "%autoreload 2\n", 73 | "%matplotlib inline\n", 74 | "!/opt/bin/nvidia-smi\n", 75 | "!nvcc --version" 76 | ], 77 | "execution_count": null, 78 | "outputs": [] 79 | }, 80 | { 81 | "cell_type": "code", 82 | "metadata": { 83 | "id": "rUlUjSarCbeu", 84 | "colab_type": "code", 85 | "colab": {} 86 | }, 87 | "source": [ 88 | "from fastai.vision import *\n", 89 | "from fastai.vision.interpret import *\n", 90 | "from fastai.callbacks.hooks import *\n", 91 | "from pathlib import Path\n", 92 | "from fastai.utils.mem import *\n", 93 | "torch.backends.cudnn.benchmark=True" 94 | ], 95 | "execution_count": null, 96 | "outputs": [] 97 | }, 98 | { 99 | "cell_type": "code", 100 | "metadata": { 101 | "id": "fbNQPOrsCdRX", 102 | "colab_type": "code", 103 | "colab": {} 104 | }, 105 | "source": [ 106 | "import warnings\n", 107 | "warnings.filterwarnings(\"ignore\", category=UserWarning, module=\"torch.nn.functional\")" 108 | ], 109 | "execution_count": null, 110 | "outputs": [] 111 | }, 112 | { 113 | "cell_type": "code", 114 | "metadata": { 115 | "id": "D3GdYMqtCexi", 116 | "colab_type": "code", 117 | "colab": {} 118 | }, 119 | "source": [ 120 | "from google.colab import drive\n", 121 | "drive.mount('/content/gdrive')" 122 | ], 123 | "execution_count": null, 124 | "outputs": [] 125 | }, 126 | { 127 | "cell_type": "markdown", 128 | "metadata": { 129 | "id": "nBWj9_LSC2G5", 130 | "colab_type": "text" 131 | }, 132 | "source": [ 133 | "## 2. Preparing the data\n", 134 | "\n", 135 | "---\n", 136 | "\n" 137 | ] 138 | }, 139 | { 140 | "cell_type": "code", 141 | "metadata": { 142 | "id": "6yMPQ7v7CtbG", 143 | "colab_type": "code", 144 | "colab": {} 145 | }, 146 | "source": [ 147 | "path = Path('gdrive/My Drive/Colab Notebooks/data/')\n", 148 | "path.ls()" 149 | ], 150 | "execution_count": null, 151 | "outputs": [] 152 | }, 153 | { 154 | "cell_type": "code", 155 | "metadata": { 156 | "id": "Z1BpN93iC8J7", 157 | "colab_type": "code", 158 | "colab": {} 159 | }, 160 | "source": [ 161 | "codes = np.loadtxt(path/'codes.txt', dtype=str); codes" 162 | ], 163 | "execution_count": null, 164 | "outputs": [] 165 | }, 166 | { 167 | "cell_type": "code", 168 | "metadata": { 169 | "id": "kfT8JAUYC-tp", 170 | "colab_type": "code", 171 | "colab": {} 172 | }, 173 | "source": [ 174 | "path_lbl = path/'labels'\n", 175 | "path_img = path/'images'" 176 | ], 177 | "execution_count": null, 178 | "outputs": [] 179 | }, 180 | { 181 | "cell_type": "code", 182 | "metadata": { 183 | "id": "N83tmLjBDAII", 184 | "colab_type": "code", 185 | "colab": {} 186 | }, 187 | "source": [ 188 | "fnames = get_image_files(path_img)\n", 189 | "fnames[:3]\n", 190 | "len(fnames)" 191 | ], 192 | "execution_count": null, 193 | "outputs": [] 194 | }, 195 | { 196 | "cell_type": "code", 197 | "metadata": { 198 | "id": "ZvTFNcwxDB6C", 199 | "colab_type": "code", 200 | "colab": {} 201 | }, 202 | "source": [ 203 | "lbl_names = get_image_files(path_lbl)\n", 204 | "lbl_names[:3]\n", 205 | "len(lbl_names)" 206 | ], 207 | "execution_count": null, 208 | "outputs": [] 209 | }, 210 | { 211 | "cell_type": "code", 212 | "metadata": { 213 | "id": "Dzk-MfPYDD7p", 214 | "colab_type": "code", 215 | "colab": {} 216 | }, 217 | "source": [ 218 | "img_f = fnames[139]\n", 219 | "img = open_image(img_f)\n", 220 | "img.show(figsize=(5,5))" 221 | ], 222 | "execution_count": null, 223 | "outputs": [] 224 | }, 225 | { 226 | "cell_type": "code", 227 | "metadata": { 228 | "id": "KxBbLefJDFsd", 229 | "colab_type": "code", 230 | "colab": {} 231 | }, 232 | "source": [ 233 | "get_y_fn = lambda x: path_lbl/f'{x.stem}{x.suffix}'" 234 | ], 235 | "execution_count": null, 236 | "outputs": [] 237 | }, 238 | { 239 | "cell_type": "code", 240 | "metadata": { 241 | "id": "P73xHjqaDHI7", 242 | "colab_type": "code", 243 | "colab": {} 244 | }, 245 | "source": [ 246 | "mask = open_mask(get_y_fn(img_f))\n", 247 | "mask.show(figsize=(5,5), alpha=1)" 248 | ], 249 | "execution_count": null, 250 | "outputs": [] 251 | }, 252 | { 253 | "cell_type": "code", 254 | "metadata": { 255 | "id": "a_2SmfMCDI7n", 256 | "colab_type": "code", 257 | "colab": {} 258 | }, 259 | "source": [ 260 | "src_size = np.array(mask.shape[1:])\n", 261 | "src_size,mask.data" 262 | ], 263 | "execution_count": null, 264 | "outputs": [] 265 | }, 266 | { 267 | "cell_type": "markdown", 268 | "metadata": { 269 | "id": "NmrXBAhNf5mC", 270 | "colab_type": "text" 271 | }, 272 | "source": [ 273 | "## 3. First Step - Without weights\n", 274 | "\n", 275 | "---\n", 276 | "\n" 277 | ] 278 | }, 279 | { 280 | "cell_type": "markdown", 281 | "metadata": { 282 | "id": "UbxyIcZkDqBJ", 283 | "colab_type": "text" 284 | }, 285 | "source": [ 286 | "### 3.1. First step Datasets\n", 287 | "\n", 288 | "---\n", 289 | "\n" 290 | ] 291 | }, 292 | { 293 | "cell_type": "code", 294 | "metadata": { 295 | "id": "KoQLPTpZDZQZ", 296 | "colab_type": "code", 297 | "colab": {} 298 | }, 299 | "source": [ 300 | "size = src_size\n", 301 | "\n", 302 | "free = gpu_mem_get_free_no_cache()\n", 303 | "# the max size of bs depends on the available GPU RAM\n", 304 | "if free > 8200: bs=8\n", 305 | "else: bs=4\n", 306 | "print(f\"using bs={bs}, have {free}MB of GPU RAM free\")" 307 | ], 308 | "execution_count": null, 309 | "outputs": [] 310 | }, 311 | { 312 | "cell_type": "code", 313 | "metadata": { 314 | "id": "ypW12JvVDxMk", 315 | "colab_type": "code", 316 | "colab": {} 317 | }, 318 | "source": [ 319 | "src = (SegmentationItemList.from_folder(path_img)\n", 320 | " .split_by_fname_file('../valid.txt')\n", 321 | " .label_from_func(get_y_fn, classes=codes))" 322 | ], 323 | "execution_count": null, 324 | "outputs": [] 325 | }, 326 | { 327 | "cell_type": "code", 328 | "metadata": { 329 | "id": "-qSJBnFCDy7t", 330 | "colab_type": "code", 331 | "colab": {} 332 | }, 333 | "source": [ 334 | "data = (src.transform(get_transforms(), size=size, tfm_y=True)\n", 335 | " .databunch(bs=bs)\n", 336 | " .normalize(imagenet_stats))" 337 | ], 338 | "execution_count": null, 339 | "outputs": [] 340 | }, 341 | { 342 | "cell_type": "code", 343 | "metadata": { 344 | "id": "-HeHMvl3D0vH", 345 | "colab_type": "code", 346 | "colab": {} 347 | }, 348 | "source": [ 349 | "data.show_batch(2, figsize=(10,10))" 350 | ], 351 | "execution_count": null, 352 | "outputs": [] 353 | }, 354 | { 355 | "cell_type": "code", 356 | "metadata": { 357 | "id": "eFnPc-sND2fW", 358 | "colab_type": "code", 359 | "colab": {} 360 | }, 361 | "source": [ 362 | "data.show_batch(2, figsize=(10,7), ds_type=DatasetType.Valid)" 363 | ], 364 | "execution_count": null, 365 | "outputs": [] 366 | }, 367 | { 368 | "cell_type": "markdown", 369 | "metadata": { 370 | "id": "K7W20An-GnNm", 371 | "colab_type": "text" 372 | }, 373 | "source": [ 374 | "### 3.2. First step Model\n", 375 | "\n", 376 | "---\n", 377 | "\n" 378 | ] 379 | }, 380 | { 381 | "cell_type": "code", 382 | "metadata": { 383 | "id": "V-paASzNELtK", 384 | "colab_type": "code", 385 | "colab": {} 386 | }, 387 | "source": [ 388 | "name2id = {v:k for k,v in enumerate(codes)}\n", 389 | "\n", 390 | "def acc_rtk(input, target):\n", 391 | " target = target.squeeze(1)\n", 392 | " mask = target != 0\n", 393 | " return (input.argmax(dim=1)[mask]==target[mask]).float().mean()" 394 | ], 395 | "execution_count": null, 396 | "outputs": [] 397 | }, 398 | { 399 | "cell_type": "code", 400 | "metadata": { 401 | "id": "NcAvDEryGuwq", 402 | "colab_type": "code", 403 | "colab": {} 404 | }, 405 | "source": [ 406 | "metrics=acc_rtk\n", 407 | "wd=1e-2" 408 | ], 409 | "execution_count": null, 410 | "outputs": [] 411 | }, 412 | { 413 | "cell_type": "code", 414 | "metadata": { 415 | "id": "F9rbvQ8AGw0a", 416 | "colab_type": "code", 417 | "colab": {} 418 | }, 419 | "source": [ 420 | "learn = unet_learner(data, models.resnet34, metrics=metrics, wd=wd)" 421 | ], 422 | "execution_count": null, 423 | "outputs": [] 424 | }, 425 | { 426 | "cell_type": "code", 427 | "metadata": { 428 | "id": "wpjCmT8mGy1W", 429 | "colab_type": "code", 430 | "colab": {} 431 | }, 432 | "source": [ 433 | "#CUDA_LAUNCH_BLOCKING=1\n", 434 | "lr_find(learn)\n", 435 | "learn.recorder.plot()" 436 | ], 437 | "execution_count": null, 438 | "outputs": [] 439 | }, 440 | { 441 | "cell_type": "code", 442 | "metadata": { 443 | "id": "GS7-jnS0G1DM", 444 | "colab_type": "code", 445 | "colab": {} 446 | }, 447 | "source": [ 448 | "lr=1e-4" 449 | ], 450 | "execution_count": null, 451 | "outputs": [] 452 | }, 453 | { 454 | "cell_type": "code", 455 | "metadata": { 456 | "id": "u0Y52M-qG3ID", 457 | "colab_type": "code", 458 | "colab": {} 459 | }, 460 | "source": [ 461 | "learn.fit_one_cycle(10, slice(lr), pct_start=0.9)" 462 | ], 463 | "execution_count": null, 464 | "outputs": [] 465 | }, 466 | { 467 | "cell_type": "code", 468 | "metadata": { 469 | "id": "Ym9mEJVVG4zE", 470 | "colab_type": "code", 471 | "colab": {} 472 | }, 473 | "source": [ 474 | "learn.save('stage-1')" 475 | ], 476 | "execution_count": null, 477 | "outputs": [] 478 | }, 479 | { 480 | "cell_type": "code", 481 | "metadata": { 482 | "id": "r_WCYmq3G6Uf", 483 | "colab_type": "code", 484 | "colab": {} 485 | }, 486 | "source": [ 487 | "learn.load('stage-1');" 488 | ], 489 | "execution_count": null, 490 | "outputs": [] 491 | }, 492 | { 493 | "cell_type": "code", 494 | "metadata": { 495 | "id": "Gb-EFFVpG7-q", 496 | "colab_type": "code", 497 | "colab": {} 498 | }, 499 | "source": [ 500 | "learn.show_results(rows=5, figsize=(15,15))" 501 | ], 502 | "execution_count": null, 503 | "outputs": [] 504 | }, 505 | { 506 | "cell_type": "markdown", 507 | "metadata": { 508 | "id": "VE4zF162G-4z", 509 | "colab_type": "text" 510 | }, 511 | "source": [ 512 | "### 3.3. Interpret\n", 513 | "\n", 514 | "---\n", 515 | "\n" 516 | ] 517 | }, 518 | { 519 | "cell_type": "code", 520 | "metadata": { 521 | "id": "kJu9vBlfHA5B", 522 | "colab_type": "code", 523 | "colab": {} 524 | }, 525 | "source": [ 526 | "interp = SegmentationInterpretation.from_learner(learn)" 527 | ], 528 | "execution_count": null, 529 | "outputs": [] 530 | }, 531 | { 532 | "cell_type": "code", 533 | "metadata": { 534 | "id": "rea5Nz9yHCxp", 535 | "colab_type": "code", 536 | "colab": {} 537 | }, 538 | "source": [ 539 | "top_losses, top_idxs = interp.top_losses((288,352))" 540 | ], 541 | "execution_count": null, 542 | "outputs": [] 543 | }, 544 | { 545 | "cell_type": "code", 546 | "metadata": { 547 | "id": "VSCFhKzzHEfS", 548 | "colab_type": "code", 549 | "colab": {} 550 | }, 551 | "source": [ 552 | "# plot loss distribution\n", 553 | "plt.hist(to_np(top_losses), bins=20)" 554 | ], 555 | "execution_count": null, 556 | "outputs": [] 557 | }, 558 | { 559 | "cell_type": "code", 560 | "metadata": { 561 | "id": "_MHIJS_yHGa7", 562 | "colab_type": "code", 563 | "colab": {} 564 | }, 565 | "source": [ 566 | "# top loss idxs of images\n", 567 | "top_idxs[:5]" 568 | ], 569 | "execution_count": null, 570 | "outputs": [] 571 | }, 572 | { 573 | "cell_type": "markdown", 574 | "metadata": { 575 | "id": "tRZoEiB8HJhI", 576 | "colab_type": "text" 577 | }, 578 | "source": [ 579 | "### 3.4. Confusion Matrix\n", 580 | "\n", 581 | "---\n", 582 | "\n" 583 | ] 584 | }, 585 | { 586 | "cell_type": "code", 587 | "metadata": { 588 | "id": "mRo-rEthHJ6A", 589 | "colab_type": "code", 590 | "colab": {} 591 | }, 592 | "source": [ 593 | "mean_cm, single_img_cm = interp._generate_confusion()" 594 | ], 595 | "execution_count": null, 596 | "outputs": [] 597 | }, 598 | { 599 | "cell_type": "code", 600 | "metadata": { 601 | "id": "mTum9BXpHNBn", 602 | "colab_type": "code", 603 | "colab": {} 604 | }, 605 | "source": [ 606 | "mean_cm.shape, single_img_cm.shape" 607 | ], 608 | "execution_count": null, 609 | "outputs": [] 610 | }, 611 | { 612 | "cell_type": "code", 613 | "metadata": { 614 | "id": "9uFc-7KcHO2l", 615 | "colab_type": "code", 616 | "colab": {} 617 | }, 618 | "source": [ 619 | "# global class performance\n", 620 | "df = interp._plot_intersect_cm(mean_cm, \"Mean of Ratio of Intersection given True Label\")" 621 | ], 622 | "execution_count": null, 623 | "outputs": [] 624 | }, 625 | { 626 | "cell_type": "code", 627 | "metadata": { 628 | "id": "D4bA9Qw4HXEB", 629 | "colab_type": "code", 630 | "colab": {} 631 | }, 632 | "source": [ 633 | "# single image class performance\n", 634 | "i = 10\n", 635 | "df = interp._plot_intersect_cm(single_img_cm[i], f\"Ratio of Intersection given True Label, Image:{i}\")" 636 | ], 637 | "execution_count": null, 638 | "outputs": [] 639 | }, 640 | { 641 | "cell_type": "code", 642 | "metadata": { 643 | "id": "tbFVrVCVHaK6", 644 | "colab_type": "code", 645 | "colab": {} 646 | }, 647 | "source": [ 648 | "# show xyz\n", 649 | "interp.show_xyz(i)" 650 | ], 651 | "execution_count": null, 652 | "outputs": [] 653 | }, 654 | { 655 | "cell_type": "markdown", 656 | "metadata": { 657 | "id": "3oW77qUxHhkp", 658 | "colab_type": "text" 659 | }, 660 | "source": [ 661 | "### 3.5. First model continuation\n", 662 | "\n", 663 | "---\n", 664 | "\n" 665 | ] 666 | }, 667 | { 668 | "cell_type": "code", 669 | "metadata": { 670 | "id": "DHhM3YfjHjmV", 671 | "colab_type": "code", 672 | "colab": {} 673 | }, 674 | "source": [ 675 | "learn.unfreeze()" 676 | ], 677 | "execution_count": null, 678 | "outputs": [] 679 | }, 680 | { 681 | "cell_type": "code", 682 | "metadata": { 683 | "id": "vpf9gZYgHlnI", 684 | "colab_type": "code", 685 | "colab": {} 686 | }, 687 | "source": [ 688 | "lrs = slice(lr/400,lr/4)" 689 | ], 690 | "execution_count": null, 691 | "outputs": [] 692 | }, 693 | { 694 | "cell_type": "code", 695 | "metadata": { 696 | "id": "h84SZCadHnaY", 697 | "colab_type": "code", 698 | "colab": {} 699 | }, 700 | "source": [ 701 | "learn.fit_one_cycle(100, lrs, pct_start=0.9)" 702 | ], 703 | "execution_count": null, 704 | "outputs": [] 705 | }, 706 | { 707 | "cell_type": "code", 708 | "metadata": { 709 | "id": "iJxgf68WHqD8", 710 | "colab_type": "code", 711 | "colab": {} 712 | }, 713 | "source": [ 714 | "learn.save('stage-2')" 715 | ], 716 | "execution_count": null, 717 | "outputs": [] 718 | }, 719 | { 720 | "cell_type": "code", 721 | "metadata": { 722 | "id": "t_rkQMWVHrmf", 723 | "colab_type": "code", 724 | "colab": {} 725 | }, 726 | "source": [ 727 | "learn.load('stage-2');" 728 | ], 729 | "execution_count": null, 730 | "outputs": [] 731 | }, 732 | { 733 | "cell_type": "code", 734 | "metadata": { 735 | "id": "rwQbJFU2HtMx", 736 | "colab_type": "code", 737 | "colab": {} 738 | }, 739 | "source": [ 740 | "learn.show_results(rows=25, figsize=(20,20))" 741 | ], 742 | "execution_count": null, 743 | "outputs": [] 744 | }, 745 | { 746 | "cell_type": "markdown", 747 | "metadata": { 748 | "id": "3ojFnvRBHx8v", 749 | "colab_type": "text" 750 | }, 751 | "source": [ 752 | "### 3.6. Confusion Matrix\n", 753 | "\n", 754 | "---\n", 755 | "\n" 756 | ] 757 | }, 758 | { 759 | "cell_type": "code", 760 | "metadata": { 761 | "id": "Sdx-NrEBiPjh", 762 | "colab_type": "code", 763 | "colab": {} 764 | }, 765 | "source": [ 766 | "interp = SegmentationInterpretation.from_learner(learn)" 767 | ], 768 | "execution_count": null, 769 | "outputs": [] 770 | }, 771 | { 772 | "cell_type": "code", 773 | "metadata": { 774 | "id": "khTXOjeCHz4U", 775 | "colab_type": "code", 776 | "colab": {} 777 | }, 778 | "source": [ 779 | "mean_cm, single_img_cm = interp._generate_confusion()" 780 | ], 781 | "execution_count": null, 782 | "outputs": [] 783 | }, 784 | { 785 | "cell_type": "code", 786 | "metadata": { 787 | "id": "LvilgGWHH2MX", 788 | "colab_type": "code", 789 | "colab": {} 790 | }, 791 | "source": [ 792 | "mean_cm.shape, single_img_cm.shape" 793 | ], 794 | "execution_count": null, 795 | "outputs": [] 796 | }, 797 | { 798 | "cell_type": "code", 799 | "metadata": { 800 | "id": "0b1RhFsxH4PW", 801 | "colab_type": "code", 802 | "colab": {} 803 | }, 804 | "source": [ 805 | "# global class performance\n", 806 | "df = interp._plot_intersect_cm(mean_cm, \"Mean of Ratio of Intersection given True Label\")" 807 | ], 808 | "execution_count": null, 809 | "outputs": [] 810 | }, 811 | { 812 | "cell_type": "code", 813 | "metadata": { 814 | "id": "XuneWk4BH64P", 815 | "colab_type": "code", 816 | "colab": {} 817 | }, 818 | "source": [ 819 | "# single image class performance\n", 820 | "i = 0\n", 821 | "df = interp._plot_intersect_cm(single_img_cm[i], f\"Ratio of Intersection given True Label, Image:{i}\")" 822 | ], 823 | "execution_count": null, 824 | "outputs": [] 825 | }, 826 | { 827 | "cell_type": "code", 828 | "metadata": { 829 | "id": "G9ZMMt2IH85B", 830 | "colab_type": "code", 831 | "colab": {} 832 | }, 833 | "source": [ 834 | "# show xyz\n", 835 | "interp.show_xyz(i)" 836 | ], 837 | "execution_count": null, 838 | "outputs": [] 839 | }, 840 | { 841 | "cell_type": "markdown", 842 | "metadata": { 843 | "id": "KgZkM8M1IAxD", 844 | "colab_type": "text" 845 | }, 846 | "source": [ 847 | "### 3.7. Interpret\n", 848 | "\n", 849 | "---\n", 850 | "\n" 851 | ] 852 | }, 853 | { 854 | "cell_type": "code", 855 | "metadata": { 856 | "id": "BXla1Fq0IBMS", 857 | "colab_type": "code", 858 | "colab": {} 859 | }, 860 | "source": [ 861 | "learn.interpret" 862 | ], 863 | "execution_count": null, 864 | "outputs": [] 865 | }, 866 | { 867 | "cell_type": "markdown", 868 | "metadata": { 869 | "id": "_siWJq89IOdJ", 870 | "colab_type": "text" 871 | }, 872 | "source": [ 873 | "### 3.8. Saving\n", 874 | "\n", 875 | "---\n", 876 | "\n" 877 | ] 878 | }, 879 | { 880 | "cell_type": "code", 881 | "metadata": { 882 | "id": "m3WfZ7bTIQOW", 883 | "colab_type": "code", 884 | "colab": {} 885 | }, 886 | "source": [ 887 | "learn.save('stage-2')" 888 | ], 889 | "execution_count": null, 890 | "outputs": [] 891 | }, 892 | { 893 | "cell_type": "code", 894 | "metadata": { 895 | "id": "HvexEcm9ISE_", 896 | "colab_type": "code", 897 | "colab": {} 898 | }, 899 | "source": [ 900 | "data=None\n", 901 | "learn=None\n", 902 | "gc.collect()" 903 | ], 904 | "execution_count": null, 905 | "outputs": [] 906 | }, 907 | { 908 | "cell_type": "markdown", 909 | "metadata": { 910 | "id": "Uh1ZPx64gR56", 911 | "colab_type": "text" 912 | }, 913 | "source": [ 914 | "## 4. Second Step - With weights\n", 915 | "\n", 916 | "---\n", 917 | "\n" 918 | ] 919 | }, 920 | { 921 | "cell_type": "markdown", 922 | "metadata": { 923 | "id": "aT8l81QTjBrE", 924 | "colab_type": "text" 925 | }, 926 | "source": [ 927 | "### 4.1. Second step Datasets\n", 928 | "\n", 929 | "---\n", 930 | "\n" 931 | ] 932 | }, 933 | { 934 | "cell_type": "code", 935 | "metadata": { 936 | "id": "FZP0o8F9i5-o", 937 | "colab_type": "code", 938 | "colab": {} 939 | }, 940 | "source": [ 941 | "size = src_size\n", 942 | "\n", 943 | "free = gpu_mem_get_free_no_cache()\n", 944 | "# the max size of bs depends on the available GPU RAM\n", 945 | "if free > 8200: bs=8\n", 946 | "else: bs=4\n", 947 | "print(f\"using bs={bs}, have {free}MB of GPU RAM free\")" 948 | ], 949 | "execution_count": null, 950 | "outputs": [] 951 | }, 952 | { 953 | "cell_type": "code", 954 | "metadata": { 955 | "id": "FgHlUXXIjMKU", 956 | "colab_type": "code", 957 | "colab": {} 958 | }, 959 | "source": [ 960 | "src = (SegmentationItemList.from_folder(path_img)\n", 961 | " .split_by_fname_file('../valid.txt')\n", 962 | " .label_from_func(get_y_fn, classes=codes))" 963 | ], 964 | "execution_count": null, 965 | "outputs": [] 966 | }, 967 | { 968 | "cell_type": "code", 969 | "metadata": { 970 | "id": "0WjndpsWjN85", 971 | "colab_type": "code", 972 | "colab": {} 973 | }, 974 | "source": [ 975 | "data = (src.transform(get_transforms(), size=size, tfm_y=True)\n", 976 | " .databunch(bs=bs)\n", 977 | " .normalize(imagenet_stats))" 978 | ], 979 | "execution_count": null, 980 | "outputs": [] 981 | }, 982 | { 983 | "cell_type": "code", 984 | "metadata": { 985 | "id": "BIaF-c96jRC4", 986 | "colab_type": "code", 987 | "colab": {} 988 | }, 989 | "source": [ 990 | "data.show_batch(2, figsize=(10,10))" 991 | ], 992 | "execution_count": null, 993 | "outputs": [] 994 | }, 995 | { 996 | "cell_type": "code", 997 | "metadata": { 998 | "id": "amJs-yS0jTj2", 999 | "colab_type": "code", 1000 | "colab": {} 1001 | }, 1002 | "source": [ 1003 | "data.show_batch(2, figsize=(10,7), ds_type=DatasetType.Valid)" 1004 | ], 1005 | "execution_count": null, 1006 | "outputs": [] 1007 | }, 1008 | { 1009 | "cell_type": "markdown", 1010 | "metadata": { 1011 | "id": "33FD1Sc2j-k0", 1012 | "colab_type": "text" 1013 | }, 1014 | "source": [ 1015 | "### 4.2. Second step Model\n", 1016 | "\n", 1017 | "---\n", 1018 | "\n" 1019 | ] 1020 | }, 1021 | { 1022 | "cell_type": "code", 1023 | "metadata": { 1024 | "id": "51MWvFIcj8PK", 1025 | "colab_type": "code", 1026 | "colab": {} 1027 | }, 1028 | "source": [ 1029 | "name2id = {v:k for k,v in enumerate(codes)}\n", 1030 | "void_code = name2id['manholeCover']\n", 1031 | "\n", 1032 | "def acc_rtk(input, target):\n", 1033 | " target = target.squeeze(1)\n", 1034 | " mask = target != void_code\n", 1035 | " return (input.argmax(dim=1)[mask]==target[mask]).float().mean()" 1036 | ], 1037 | "execution_count": null, 1038 | "outputs": [] 1039 | }, 1040 | { 1041 | "cell_type": "code", 1042 | "metadata": { 1043 | "id": "U-TFrct9kEI7", 1044 | "colab_type": "code", 1045 | "colab": {} 1046 | }, 1047 | "source": [ 1048 | "metrics=acc_rtk\n", 1049 | "wd=1e-2" 1050 | ], 1051 | "execution_count": null, 1052 | "outputs": [] 1053 | }, 1054 | { 1055 | "cell_type": "code", 1056 | "metadata": { 1057 | "id": "XhlXrkb3kGqL", 1058 | "colab_type": "code", 1059 | "colab": {} 1060 | }, 1061 | "source": [ 1062 | "balanced_loss = CrossEntropyFlat(axis=1, weight=torch.tensor([1.0,5.0,6.0,7.0,75.0,1000.0,3100.0,3300.0,0.0,270.0,2200.0,1000.0,180.0]).cuda())" 1063 | ], 1064 | "execution_count": null, 1065 | "outputs": [] 1066 | }, 1067 | { 1068 | "cell_type": "code", 1069 | "metadata": { 1070 | "id": "r53gfgy6kKsP", 1071 | "colab_type": "code", 1072 | "colab": {} 1073 | }, 1074 | "source": [ 1075 | "learn = unet_learner(data, models.resnet34, metrics=metrics, loss_func=balanced_loss, wd=wd)" 1076 | ], 1077 | "execution_count": null, 1078 | "outputs": [] 1079 | }, 1080 | { 1081 | "cell_type": "code", 1082 | "metadata": { 1083 | "id": "QZJ69o4ZkRW3", 1084 | "colab_type": "code", 1085 | "colab": {} 1086 | }, 1087 | "source": [ 1088 | "learn.load('stage-2')" 1089 | ], 1090 | "execution_count": null, 1091 | "outputs": [] 1092 | }, 1093 | { 1094 | "cell_type": "code", 1095 | "metadata": { 1096 | "id": "LdbVHXsWkVLJ", 1097 | "colab_type": "code", 1098 | "colab": {} 1099 | }, 1100 | "source": [ 1101 | "!/opt/bin/nvidia-smi" 1102 | ], 1103 | "execution_count": null, 1104 | "outputs": [] 1105 | }, 1106 | { 1107 | "cell_type": "code", 1108 | "metadata": { 1109 | "id": "BrjsLWBnkXM4", 1110 | "colab_type": "code", 1111 | "colab": {} 1112 | }, 1113 | "source": [ 1114 | "#CUDA_LAUNCH_BLOCKING=1\n", 1115 | "lr_find(learn)\n", 1116 | "learn.recorder.plot()" 1117 | ], 1118 | "execution_count": null, 1119 | "outputs": [] 1120 | }, 1121 | { 1122 | "cell_type": "code", 1123 | "metadata": { 1124 | "id": "vF7XVkIhkZIf", 1125 | "colab_type": "code", 1126 | "colab": {} 1127 | }, 1128 | "source": [ 1129 | "lr=1e-4" 1130 | ], 1131 | "execution_count": null, 1132 | "outputs": [] 1133 | }, 1134 | { 1135 | "cell_type": "code", 1136 | "metadata": { 1137 | "id": "MmG820FAkazZ", 1138 | "colab_type": "code", 1139 | "colab": {} 1140 | }, 1141 | "source": [ 1142 | "learn.fit_one_cycle(10, slice(lr), pct_start=0.9)" 1143 | ], 1144 | "execution_count": null, 1145 | "outputs": [] 1146 | }, 1147 | { 1148 | "cell_type": "code", 1149 | "metadata": { 1150 | "id": "IV2AM1tVkdNy", 1151 | "colab_type": "code", 1152 | "colab": {} 1153 | }, 1154 | "source": [ 1155 | "learn.save('stage-1-weights')" 1156 | ], 1157 | "execution_count": null, 1158 | "outputs": [] 1159 | }, 1160 | { 1161 | "cell_type": "code", 1162 | "metadata": { 1163 | "id": "cnrMAAJpkfNM", 1164 | "colab_type": "code", 1165 | "colab": {} 1166 | }, 1167 | "source": [ 1168 | "learn.load('stage-1-weights');" 1169 | ], 1170 | "execution_count": null, 1171 | "outputs": [] 1172 | }, 1173 | { 1174 | "cell_type": "code", 1175 | "metadata": { 1176 | "id": "JjrNgXZXkg9_", 1177 | "colab_type": "code", 1178 | "colab": {} 1179 | }, 1180 | "source": [ 1181 | "learn.show_results(rows=5, figsize=(15,15))" 1182 | ], 1183 | "execution_count": null, 1184 | "outputs": [] 1185 | }, 1186 | { 1187 | "cell_type": "markdown", 1188 | "metadata": { 1189 | "id": "4TAIZ9s4msq5", 1190 | "colab_type": "text" 1191 | }, 1192 | "source": [ 1193 | "### 4.3. Interpret\n", 1194 | "\n", 1195 | "---\n", 1196 | "\n" 1197 | ] 1198 | }, 1199 | { 1200 | "cell_type": "code", 1201 | "metadata": { 1202 | "id": "W74hI7tWmncC", 1203 | "colab_type": "code", 1204 | "colab": {} 1205 | }, 1206 | "source": [ 1207 | "interp = SegmentationInterpretation.from_learner(learn)" 1208 | ], 1209 | "execution_count": null, 1210 | "outputs": [] 1211 | }, 1212 | { 1213 | "cell_type": "code", 1214 | "metadata": { 1215 | "id": "yel6fEhlmxwA", 1216 | "colab_type": "code", 1217 | "colab": {} 1218 | }, 1219 | "source": [ 1220 | "top_losses, top_idxs = interp.top_losses((288,352))" 1221 | ], 1222 | "execution_count": null, 1223 | "outputs": [] 1224 | }, 1225 | { 1226 | "cell_type": "code", 1227 | "metadata": { 1228 | "id": "6yfQcvPvmzd6", 1229 | "colab_type": "code", 1230 | "colab": {} 1231 | }, 1232 | "source": [ 1233 | "# plot loss distribution\n", 1234 | "plt.hist(to_np(top_losses), bins=20)" 1235 | ], 1236 | "execution_count": null, 1237 | "outputs": [] 1238 | }, 1239 | { 1240 | "cell_type": "code", 1241 | "metadata": { 1242 | "id": "dVEVTr0Vm2cj", 1243 | "colab_type": "code", 1244 | "colab": {} 1245 | }, 1246 | "source": [ 1247 | "# top loss idxs of images\n", 1248 | "top_idxs[:5]" 1249 | ], 1250 | "execution_count": null, 1251 | "outputs": [] 1252 | }, 1253 | { 1254 | "cell_type": "markdown", 1255 | "metadata": { 1256 | "id": "D8N5E7BxnK7n", 1257 | "colab_type": "text" 1258 | }, 1259 | "source": [ 1260 | "### 4.4. Confusion Matrix\n", 1261 | "\n", 1262 | "---\n", 1263 | "\n" 1264 | ] 1265 | }, 1266 | { 1267 | "cell_type": "code", 1268 | "metadata": { 1269 | "id": "WzpLXsjAnCzg", 1270 | "colab_type": "code", 1271 | "colab": {} 1272 | }, 1273 | "source": [ 1274 | "mean_cm, single_img_cm = interp._generate_confusion()" 1275 | ], 1276 | "execution_count": null, 1277 | "outputs": [] 1278 | }, 1279 | { 1280 | "cell_type": "code", 1281 | "metadata": { 1282 | "id": "qlGEqxtKnPcH", 1283 | "colab_type": "code", 1284 | "colab": {} 1285 | }, 1286 | "source": [ 1287 | "mean_cm.shape, single_img_cm.shape" 1288 | ], 1289 | "execution_count": null, 1290 | "outputs": [] 1291 | }, 1292 | { 1293 | "cell_type": "code", 1294 | "metadata": { 1295 | "id": "ypwBccNlnRxJ", 1296 | "colab_type": "code", 1297 | "colab": {} 1298 | }, 1299 | "source": [ 1300 | "# global class performance\n", 1301 | "df = interp._plot_intersect_cm(mean_cm, \"Mean of Ratio of Intersection given True Label\")" 1302 | ], 1303 | "execution_count": null, 1304 | "outputs": [] 1305 | }, 1306 | { 1307 | "cell_type": "code", 1308 | "metadata": { 1309 | "id": "_4CNorVgnU2D", 1310 | "colab_type": "code", 1311 | "colab": {} 1312 | }, 1313 | "source": [ 1314 | "# single image class performance\n", 1315 | "i = 10\n", 1316 | "df = interp._plot_intersect_cm(single_img_cm[i], f\"Ratio of Intersection given True Label, Image:{i}\")" 1317 | ], 1318 | "execution_count": null, 1319 | "outputs": [] 1320 | }, 1321 | { 1322 | "cell_type": "code", 1323 | "metadata": { 1324 | "id": "V2h5ich-nXmo", 1325 | "colab_type": "code", 1326 | "colab": {} 1327 | }, 1328 | "source": [ 1329 | "# show xyz\n", 1330 | "interp.show_xyz(i)" 1331 | ], 1332 | "execution_count": null, 1333 | "outputs": [] 1334 | }, 1335 | { 1336 | "cell_type": "markdown", 1337 | "metadata": { 1338 | "id": "XI-bgt6voXi9", 1339 | "colab_type": "text" 1340 | }, 1341 | "source": [ 1342 | "### 4.5. Second model continuation\n", 1343 | "\n", 1344 | "---\n", 1345 | "\n" 1346 | ] 1347 | }, 1348 | { 1349 | "cell_type": "code", 1350 | "metadata": { 1351 | "id": "6BuzgDOYoS60", 1352 | "colab_type": "code", 1353 | "colab": {} 1354 | }, 1355 | "source": [ 1356 | "learn.unfreeze()" 1357 | ], 1358 | "execution_count": null, 1359 | "outputs": [] 1360 | }, 1361 | { 1362 | "cell_type": "code", 1363 | "metadata": { 1364 | "id": "Zili8A1rojNi", 1365 | "colab_type": "code", 1366 | "colab": {} 1367 | }, 1368 | "source": [ 1369 | "lrs = slice(lr/400,lr/4)" 1370 | ], 1371 | "execution_count": null, 1372 | "outputs": [] 1373 | }, 1374 | { 1375 | "cell_type": "code", 1376 | "metadata": { 1377 | "id": "6C5gFMv7olCX", 1378 | "colab_type": "code", 1379 | "colab": {} 1380 | }, 1381 | "source": [ 1382 | "learn.fit_one_cycle(100, lrs, pct_start=0.8)" 1383 | ], 1384 | "execution_count": null, 1385 | "outputs": [] 1386 | }, 1387 | { 1388 | "cell_type": "code", 1389 | "metadata": { 1390 | "id": "DwdGGDd3ooXx", 1391 | "colab_type": "code", 1392 | "colab": {} 1393 | }, 1394 | "source": [ 1395 | "learn.save('stage-2-weights')" 1396 | ], 1397 | "execution_count": null, 1398 | "outputs": [] 1399 | }, 1400 | { 1401 | "cell_type": "code", 1402 | "metadata": { 1403 | "id": "Oa5iJP78osiY", 1404 | "colab_type": "code", 1405 | "colab": {} 1406 | }, 1407 | "source": [ 1408 | "learn.load('stage-2-weights');" 1409 | ], 1410 | "execution_count": null, 1411 | "outputs": [] 1412 | }, 1413 | { 1414 | "cell_type": "code", 1415 | "metadata": { 1416 | "id": "N4yDmcE7ouv6", 1417 | "colab_type": "code", 1418 | "colab": {} 1419 | }, 1420 | "source": [ 1421 | "learn.show_results(rows=25, figsize=(20,20))" 1422 | ], 1423 | "execution_count": null, 1424 | "outputs": [] 1425 | }, 1426 | { 1427 | "cell_type": "markdown", 1428 | "metadata": { 1429 | "id": "wAZbjvb40iMJ", 1430 | "colab_type": "text" 1431 | }, 1432 | "source": [ 1433 | "### 4.6. Confusion Matrix\n", 1434 | "\n", 1435 | "---\n", 1436 | "\n" 1437 | ] 1438 | }, 1439 | { 1440 | "cell_type": "code", 1441 | "metadata": { 1442 | "id": "V1zqX1rEicia", 1443 | "colab_type": "code", 1444 | "colab": {} 1445 | }, 1446 | "source": [ 1447 | "interp = SegmentationInterpretation.from_learner(learn)" 1448 | ], 1449 | "execution_count": null, 1450 | "outputs": [] 1451 | }, 1452 | { 1453 | "cell_type": "code", 1454 | "metadata": { 1455 | "id": "s4K_t8a30QW4", 1456 | "colab_type": "code", 1457 | "colab": {} 1458 | }, 1459 | "source": [ 1460 | "mean_cm, single_img_cm = interp._generate_confusion()" 1461 | ], 1462 | "execution_count": null, 1463 | "outputs": [] 1464 | }, 1465 | { 1466 | "cell_type": "code", 1467 | "metadata": { 1468 | "id": "P2jgTPz60mMm", 1469 | "colab_type": "code", 1470 | "colab": {} 1471 | }, 1472 | "source": [ 1473 | "mean_cm.shape, single_img_cm.shape" 1474 | ], 1475 | "execution_count": null, 1476 | "outputs": [] 1477 | }, 1478 | { 1479 | "cell_type": "code", 1480 | "metadata": { 1481 | "id": "k149KJTd0oCk", 1482 | "colab_type": "code", 1483 | "colab": {} 1484 | }, 1485 | "source": [ 1486 | "# global class performance\n", 1487 | "df = interp._plot_intersect_cm(mean_cm, \"Mean of Ratio of Intersection given True Label\")" 1488 | ], 1489 | "execution_count": null, 1490 | "outputs": [] 1491 | }, 1492 | { 1493 | "cell_type": "code", 1494 | "metadata": { 1495 | "id": "ccTzua6t0q3z", 1496 | "colab_type": "code", 1497 | "colab": {} 1498 | }, 1499 | "source": [ 1500 | "# single image class performance\n", 1501 | "i = 130\n", 1502 | "df = interp._plot_intersect_cm(single_img_cm[i], f\"Ratio of Intersection given True Label, Image:{i}\")" 1503 | ], 1504 | "execution_count": null, 1505 | "outputs": [] 1506 | }, 1507 | { 1508 | "cell_type": "code", 1509 | "metadata": { 1510 | "id": "StyO909Y0tbq", 1511 | "colab_type": "code", 1512 | "colab": {} 1513 | }, 1514 | "source": [ 1515 | "# show xyz\n", 1516 | "interp.show_xyz(i)" 1517 | ], 1518 | "execution_count": null, 1519 | "outputs": [] 1520 | }, 1521 | { 1522 | "cell_type": "markdown", 1523 | "metadata": { 1524 | "id": "g5RVRyXL0_Uh", 1525 | "colab_type": "text" 1526 | }, 1527 | "source": [ 1528 | "### 4.7. Interpret\n", 1529 | "\n", 1530 | "---\n", 1531 | "\n" 1532 | ] 1533 | }, 1534 | { 1535 | "cell_type": "code", 1536 | "metadata": { 1537 | "id": "DHSOVNYk0_yI", 1538 | "colab_type": "code", 1539 | "colab": {} 1540 | }, 1541 | "source": [ 1542 | "learn.interpret" 1543 | ], 1544 | "execution_count": null, 1545 | "outputs": [] 1546 | }, 1547 | { 1548 | "cell_type": "markdown", 1549 | "metadata": { 1550 | "id": "eZ7SZZBN1KrS", 1551 | "colab_type": "text" 1552 | }, 1553 | "source": [ 1554 | "### 4.8. Saving\n", 1555 | "\n", 1556 | "---\n", 1557 | "\n" 1558 | ] 1559 | }, 1560 | { 1561 | "cell_type": "code", 1562 | "metadata": { 1563 | "id": "L-jfTKtS1GwG", 1564 | "colab_type": "code", 1565 | "colab": {} 1566 | }, 1567 | "source": [ 1568 | "learn.save('stage-2-weights')" 1569 | ], 1570 | "execution_count": null, 1571 | "outputs": [] 1572 | }, 1573 | { 1574 | "cell_type": "markdown", 1575 | "metadata": { 1576 | "id": "S0FHwE1p1RmR", 1577 | "colab_type": "text" 1578 | }, 1579 | "source": [ 1580 | "## 5. Using a saved model to Predict\n", 1581 | "\n", 1582 | "---\n", 1583 | "\n" 1584 | ] 1585 | }, 1586 | { 1587 | "cell_type": "code", 1588 | "metadata": { 1589 | "id": "zs0LgfQr1NQ1", 1590 | "colab_type": "code", 1591 | "colab": {} 1592 | }, 1593 | "source": [ 1594 | "size = src_size\n", 1595 | "\n", 1596 | "free = gpu_mem_get_free_no_cache()\n", 1597 | "# the max size of bs depends on the available GPU RAM\n", 1598 | "if free > 8200: bs=8\n", 1599 | "else: bs=4\n", 1600 | "print(f\"using bs={bs}, have {free}MB of GPU RAM free\")" 1601 | ], 1602 | "execution_count": null, 1603 | "outputs": [] 1604 | }, 1605 | { 1606 | "cell_type": "code", 1607 | "metadata": { 1608 | "id": "kcPVDg1o1ZyY", 1609 | "colab_type": "code", 1610 | "colab": {} 1611 | }, 1612 | "source": [ 1613 | "src = (SegmentationItemList.from_folder(path_img)\n", 1614 | " .split_by_fname_file('../valid.txt')\n", 1615 | " .label_from_func(get_y_fn, classes=codes))" 1616 | ], 1617 | "execution_count": null, 1618 | "outputs": [] 1619 | }, 1620 | { 1621 | "cell_type": "code", 1622 | "metadata": { 1623 | "id": "Xhppsx_R1bvI", 1624 | "colab_type": "code", 1625 | "colab": {} 1626 | }, 1627 | "source": [ 1628 | "data = (src.transform(get_transforms(), size=size, tfm_y=True)\n", 1629 | " .databunch(bs=bs)\n", 1630 | " .normalize(imagenet_stats))" 1631 | ], 1632 | "execution_count": null, 1633 | "outputs": [] 1634 | }, 1635 | { 1636 | "cell_type": "code", 1637 | "metadata": { 1638 | "id": "ViZg6ouZOdcm", 1639 | "colab_type": "code", 1640 | "colab": {} 1641 | }, 1642 | "source": [ 1643 | "learn = unet_learner(data, models.resnet34)" 1644 | ], 1645 | "execution_count": null, 1646 | "outputs": [] 1647 | }, 1648 | { 1649 | "cell_type": "code", 1650 | "metadata": { 1651 | "id": "wkRfgXLd1sMe", 1652 | "colab_type": "code", 1653 | "colab": {} 1654 | }, 1655 | "source": [ 1656 | "learn.load('stage-2-weights');" 1657 | ], 1658 | "execution_count": null, 1659 | "outputs": [] 1660 | }, 1661 | { 1662 | "cell_type": "code", 1663 | "metadata": { 1664 | "id": "ABRotkkd1vSU", 1665 | "colab_type": "code", 1666 | "colab": {} 1667 | }, 1668 | "source": [ 1669 | "interp = SegmentationInterpretation.from_learner(learn)" 1670 | ], 1671 | "execution_count": null, 1672 | "outputs": [] 1673 | }, 1674 | { 1675 | "cell_type": "code", 1676 | "metadata": { 1677 | "id": "gE8c7YSX1xOH", 1678 | "colab_type": "code", 1679 | "colab": {} 1680 | }, 1681 | "source": [ 1682 | "mean_cm, single_img_cm = interp._generate_confusion()" 1683 | ], 1684 | "execution_count": null, 1685 | "outputs": [] 1686 | }, 1687 | { 1688 | "cell_type": "code", 1689 | "metadata": { 1690 | "id": "OWUm7CsA1yxT", 1691 | "colab_type": "code", 1692 | "colab": {} 1693 | }, 1694 | "source": [ 1695 | "mean_cm.shape, single_img_cm.shape" 1696 | ], 1697 | "execution_count": null, 1698 | "outputs": [] 1699 | }, 1700 | { 1701 | "cell_type": "code", 1702 | "metadata": { 1703 | "id": "6ZejuHxi106m", 1704 | "colab_type": "code", 1705 | "colab": {} 1706 | }, 1707 | "source": [ 1708 | "# global class performance\n", 1709 | "df = interp._plot_intersect_cm(mean_cm, \"Mean of Ratio of Intersection given True Label\")" 1710 | ], 1711 | "execution_count": null, 1712 | "outputs": [] 1713 | }, 1714 | { 1715 | "cell_type": "code", 1716 | "metadata": { 1717 | "id": "vGajg-Q_13xU", 1718 | "colab_type": "code", 1719 | "colab": {} 1720 | }, 1721 | "source": [ 1722 | "# single image class performance\n", 1723 | "i = 130\n", 1724 | "df = interp._plot_intersect_cm(single_img_cm[i], f\"Ratio of Intersection given True Label, Image:{i}\")" 1725 | ], 1726 | "execution_count": null, 1727 | "outputs": [] 1728 | }, 1729 | { 1730 | "cell_type": "code", 1731 | "metadata": { 1732 | "id": "KmLQEz2w16H6", 1733 | "colab_type": "code", 1734 | "colab": {} 1735 | }, 1736 | "source": [ 1737 | "# show xyz\n", 1738 | "interp.show_xyz(i)" 1739 | ], 1740 | "execution_count": null, 1741 | "outputs": [] 1742 | }, 1743 | { 1744 | "cell_type": "code", 1745 | "metadata": { 1746 | "id": "J66cvtAP181a", 1747 | "colab_type": "code", 1748 | "colab": {} 1749 | }, 1750 | "source": [ 1751 | "learn.show_results()" 1752 | ], 1753 | "execution_count": null, 1754 | "outputs": [] 1755 | }, 1756 | { 1757 | "cell_type": "markdown", 1758 | "metadata": { 1759 | "id": "IO3-KXDsp4VB", 1760 | "colab_type": "text" 1761 | }, 1762 | "source": [ 1763 | "## 6. Saving the Results\n", 1764 | "\n", 1765 | "---\n", 1766 | "\n" 1767 | ] 1768 | }, 1769 | { 1770 | "cell_type": "code", 1771 | "metadata": { 1772 | "id": "BMj7IsIep6oZ", 1773 | "colab_type": "code", 1774 | "colab": {} 1775 | }, 1776 | "source": [ 1777 | "img_f = fnames[655]\n", 1778 | "img = open_image(img_f)\n", 1779 | "img.show(figsize=(5,5))" 1780 | ], 1781 | "execution_count": null, 1782 | "outputs": [] 1783 | }, 1784 | { 1785 | "cell_type": "code", 1786 | "metadata": { 1787 | "id": "m5j5g8nuqCd5", 1788 | "colab_type": "code", 1789 | "colab": {} 1790 | }, 1791 | "source": [ 1792 | "prediction = learn.predict(img)" 1793 | ], 1794 | "execution_count": null, 1795 | "outputs": [] 1796 | }, 1797 | { 1798 | "cell_type": "code", 1799 | "metadata": { 1800 | "id": "r8F26wXwqEIL", 1801 | "colab_type": "code", 1802 | "colab": {} 1803 | }, 1804 | "source": [ 1805 | "prediction[0].show(figsize=(5,5))" 1806 | ], 1807 | "execution_count": null, 1808 | "outputs": [] 1809 | }, 1810 | { 1811 | "cell_type": "code", 1812 | "metadata": { 1813 | "id": "S47GvGCLqGTe", 1814 | "colab_type": "code", 1815 | "colab": {} 1816 | }, 1817 | "source": [ 1818 | "results_save = 'results'\n", 1819 | "path_rst = path/results_save\n", 1820 | "path_rst.mkdir(exist_ok=True)" 1821 | ], 1822 | "execution_count": null, 1823 | "outputs": [] 1824 | }, 1825 | { 1826 | "cell_type": "code", 1827 | "metadata": { 1828 | "id": "J8KKlUcSqLZ1", 1829 | "colab_type": "code", 1830 | "colab": {} 1831 | }, 1832 | "source": [ 1833 | "def save_preds(names):\n", 1834 | " i=0\n", 1835 | " #names = dl.dataset.items\n", 1836 | " \n", 1837 | " for b in names:\n", 1838 | " img_s = fnames[i]\n", 1839 | " img_toSave = open_image(img_s)\n", 1840 | " img_split = f'{img_s}'\n", 1841 | " img_split = img_split[44:]\n", 1842 | " predictionSave = learn.predict(img_toSave)\n", 1843 | " predictionSave[0].save(path_rst/img_split) #Save Image\n", 1844 | " i += 1\n", 1845 | " print(i)" 1846 | ], 1847 | "execution_count": null, 1848 | "outputs": [] 1849 | }, 1850 | { 1851 | "cell_type": "code", 1852 | "metadata": { 1853 | "id": "sGOALMQJqNT0", 1854 | "colab_type": "code", 1855 | "colab": {} 1856 | }, 1857 | "source": [ 1858 | "save_preds(fnames)" 1859 | ], 1860 | "execution_count": null, 1861 | "outputs": [] 1862 | }, 1863 | { 1864 | "cell_type": "markdown", 1865 | "metadata": { 1866 | "id": "Ke8FATsZW1n_", 1867 | "colab_type": "text" 1868 | }, 1869 | "source": [ 1870 | "## 7. Coloring the Results\n", 1871 | "\n", 1872 | "---\n", 1873 | "\n", 1874 | "\n" 1875 | ] 1876 | }, 1877 | { 1878 | "cell_type": "code", 1879 | "metadata": { 1880 | "id": "8ODnS4N8W6yF", 1881 | "colab_type": "code", 1882 | "colab": {} 1883 | }, 1884 | "source": [ 1885 | "import os\n", 1886 | "import glob\n", 1887 | "import base64\n", 1888 | "import cv2 as cv\n", 1889 | "cv.__version__" 1890 | ], 1891 | "execution_count": null, 1892 | "outputs": [] 1893 | }, 1894 | { 1895 | "cell_type": "code", 1896 | "metadata": { 1897 | "id": "H5ZL2LkXX9Ij", 1898 | "colab_type": "code", 1899 | "colab": {} 1900 | }, 1901 | "source": [ 1902 | "colored_results = 'results_color'\n", 1903 | "path_crst = path/colored_results\n", 1904 | "path_crst.mkdir(exist_ok=True)" 1905 | ], 1906 | "execution_count": null, 1907 | "outputs": [] 1908 | }, 1909 | { 1910 | "cell_type": "code", 1911 | "metadata": { 1912 | "id": "CaMwRbGJzFyG", 1913 | "colab_type": "code", 1914 | "colab": {} 1915 | }, 1916 | "source": [ 1917 | "%load_ext cython" 1918 | ], 1919 | "execution_count": null, 1920 | "outputs": [] 1921 | }, 1922 | { 1923 | "cell_type": "code", 1924 | "metadata": { 1925 | "id": "5tdvtka_zJg7", 1926 | "colab_type": "code", 1927 | "colab": {} 1928 | }, 1929 | "source": [ 1930 | "%%cython -a\n", 1931 | "import cython\n", 1932 | "cimport numpy\n", 1933 | "import cv2 as cv\n", 1934 | "import numpy as np\n", 1935 | "\n", 1936 | "@cython.boundscheck(False)\n", 1937 | "@cython.wraparound(False)\n", 1938 | "#def colorfull_fast(numpy.ndarray[numpy.uint8_t, ndim=3, mode=\"c\"] frame):\n", 1939 | "cpdef numpy.ndarray[numpy.uint8_t, ndim=3, mode=\"c\"] colorfull_fast(numpy.ndarray[numpy.uint8_t, ndim=3, mode=\"c\"] frame):\n", 1940 | " # set the variable extension types\n", 1941 | " cdef int x, y, width, height, b, g, r\n", 1942 | "\n", 1943 | " #frame = cv.imdecode(np.frombuffer(byteframe, np.uint8), -1)\n", 1944 | "\n", 1945 | " # grab the image dimensions\n", 1946 | " width = 288\n", 1947 | " height = 352\n", 1948 | " \n", 1949 | " # loop over the image, pixel by pixel\n", 1950 | " for x in range(width):\n", 1951 | " for y in range(height):\n", 1952 | " b, g, r = frame[x, y]\n", 1953 | " if (b, g, r) == (0,0,0): #background\n", 1954 | " frame[x, y] = (0,0,0)\n", 1955 | " elif (b, g, r) == (1,1,1): #roadAsphalt\n", 1956 | " frame[x, y] = (85,85,255)\n", 1957 | " elif (b, g, r) == (2,2,2): #roadPaved\n", 1958 | " frame[x, y] = (85,170,127)\n", 1959 | " elif (b, g, r) == (3,3,3): #roadUnpaved\n", 1960 | " frame[x, y] = (255,170,127) \n", 1961 | " elif (b, g, r) == (4,4,4): #roadMarking\n", 1962 | " frame[x, y] = (255,255,255) \n", 1963 | " elif (b, g, r) == (5,5,5): #speedBump\n", 1964 | " frame[x, y] = (255,85,255)\n", 1965 | " elif (b, g, r) == (6,6,6): #catsEye\n", 1966 | " frame[x, y] = (255,255,127) \n", 1967 | " elif (b, g, r) == (7,7,7): #stormDrain\n", 1968 | " frame[x, y] = (170,0,127) \n", 1969 | " elif (b, g, r) == (8,8,8): #manholeCover\n", 1970 | " frame[x, y] = (0,255,255) \n", 1971 | " elif (b, g, r) == (9,9,9): #patchs\n", 1972 | " frame[x, y] = (0,0,127) \n", 1973 | " elif (b, g, r) == (10,10,10): #waterPuddle\n", 1974 | " frame[x, y] = (170,0,0)\n", 1975 | " elif (b, g, r) == (11,11,11): #pothole\n", 1976 | " frame[x, y] = (255,0,0)\n", 1977 | " elif (b, g, r) == (12,12,12): #cracks\n", 1978 | " frame[x, y] = (255,85,0)\n", 1979 | " \n", 1980 | " frame = cv.cvtColor(frame,cv.COLOR_BGR2RGB)\n", 1981 | " \n", 1982 | " # return the colored image\n", 1983 | " return frame" 1984 | ], 1985 | "execution_count": null, 1986 | "outputs": [] 1987 | }, 1988 | { 1989 | "cell_type": "code", 1990 | "metadata": { 1991 | "id": "kAidoH1KzMsd", 1992 | "colab_type": "code", 1993 | "colab": {} 1994 | }, 1995 | "source": [ 1996 | "import timeit\n", 1997 | "# Count variables\n", 1998 | "fqtd = 0\n", 1999 | "\n", 2000 | "filenames = [img for img in glob.glob(str(path_rst/\"*.png\"))]\n", 2001 | "\n", 2002 | "filenames.sort() # ADD THIS LINE\n", 2003 | "\n", 2004 | "for img in filenames:\n", 2005 | " frame = cv.imread(img)\n", 2006 | "\n", 2007 | " #%timeit colorfull_fast(frame)\n", 2008 | " \n", 2009 | " frame = colorfull_fast(frame)\n", 2010 | " name = \"%09d.png\"%fqtd\n", 2011 | " cv.imwrite(os.path.join(path_crst, name), frame)\n", 2012 | "\n", 2013 | " fqtd += 1\n", 2014 | " print(fqtd)\n", 2015 | "\n", 2016 | "print(\"Done!\")" 2017 | ], 2018 | "execution_count": null, 2019 | "outputs": [] 2020 | } 2021 | ] 2022 | } --------------------------------------------------------------------------------