├── Detectron2_AIHub_Tutorial.ipynb ├── Detectron2_Tutorial_(kor_ver).ipynb ├── Detectron2_training_custom_dataset_(AIHub_sidewalk_datataset).ipynb ├── README.md └── models ├── events.out.tfevents.1575349464.9bcafb0ea309.238.0 ├── metrics.json └── model.txt /README.md: -------------------------------------------------------------------------------- 1 | # AIHub Dataset + Detectron2 Tutorial 2 | [Detectron2](https://github.com/facebookresearch/detectron2/tree/master/detectron2)에서 제공하는 Object Detection 알고리즘을 기반으로 [AI허브](http://aihub.or.kr/content/611)에서 제공하는 보행자 데이터셋을 학습시키기 위한 튜토리얼 문서입니다. 3 | 4 | 또한, AI허브 보행자 데이터셋으로 학습된 pre-trained weight를 제공합니다. 5 | 6 | ![image](https://user-images.githubusercontent.com/15168540/70385966-8ac3b980-19d6-11ea-94c7-7c793fb810f7.png) 7 | ![image](https://user-images.githubusercontent.com/15168540/70727477-1759d900-1d43-11ea-912b-de7c8797bdce.png) 8 | ![image](https://user-images.githubusercontent.com/15168540/70727520-28a2e580-1d43-11ea-80ab-636da041f292.png) 9 | ![image](https://user-images.githubusercontent.com/15168540/70729772-22af0380-1d47-11ea-8afe-876cae83bd8b.png) 10 | 11 | 12 | **데모 비디오** 13 | [YouTube link](https://youtu.be/QpmHKVqvufo) 14 | 15 | # Tutorials 16 | 1. [Detectron2 공식 튜토리얼 번역 (pre-trained by MS-COCO)](https://colab.research.google.com/github/visionNoob/detectron2_aihub_tutorial/blob/master/Detectron2_Tutorial_(kor_ver).ipynb) 17 | 2. [Detectron2 AIHUB 데이터셋 튜토리얼(inference)](https://colab.research.google.com/github/visionNoob/detectron2_aihub_tutorial/blob/master/Detectron2_AIHub_Tutorial.ipynb) 18 | 3. [Detectron2 AIHUB 데이터셋 튜토리얼(training, jupyter notebook)](https://github.com/visionNoob/detectron2_aihub_tutorial/blob/master/Detectron2_training_custom_dataset_(AIHub_sidewalk_datataset).ipynb) 19 | 20 | # AI허브 인도 보행 데이터셋 정보 21 | [다운로드 링크](http://aihub.or.kr/content/611) 22 | 23 | - 전체 이미지 수 : 105068장 24 | - 객체 종류 : 27가지 25 | - 전체 객체 수 : 1127907 26 | 27 | 28 | ## Data Statistics 29 | |Class|#|Class|#| 30 | |-|-|-|-| 31 | |car|270978|tree_trunk|159540| 32 | |person|143667|pole|130462| 33 | |bollard|102722|traffic_sign|53463| 34 | |traffic_light|49204|truck|44319| 35 | |movable_signage|43303|potted_plant|29284| 36 | |bicycle|19507|motorcycle|18108| 37 | |bus|16020|bench|11116| 38 | |chair|10122|barricade|6005| 39 | |stop|5039|fire_hydrant|3385| 40 | |table|3312|kiosk|3293| 41 | |carrier|3083|stroller|790| 42 | |scooter|429|parking_meter|251| 43 | |dog|226|wheelchair|215| 44 | |cat|63| 45 | --- 46 | ![image](https://user-images.githubusercontent.com/15168540/70509636-5be24a80-1b72-11ea-8a4d-a4fd0594012d.png) 47 | 48 | # Trainig Model 49 | ## 1. DATASET 구성 50 | 데이터셋을 다음과 같이 구성하세요. 51 | 52 | 다운로드 53 | [aihub_27_classes_label.csv](https://www.dropbox.com/s/mh3yeeu22h8spqb/aihub_27_classes_label.csv 54 | ) 55 | [aihub_13_classes_label.csv](https://www.dropbox.com/s/byecey0zebrn203/aihub_13_classes_label.csv 56 | ) 57 | 58 | ``` 59 | ${DATASET_ROOT} 60 | `-- aihub_27_classes_label.csv 61 | -- aihub_13_classes_label.csv 62 | -- images` 63 | |-- 0617_01 64 | | |-- 0617_01.xml 65 | | |-- MP_SEL_000001.jpg 66 | | |-- MP_SEL_000002.jpg 67 | | |-- MP_SEL_000003.jpg 68 | | |-- MP_SEL_000004.jpg 69 | | |-- ... 70 | |-- 0617_02 71 | | |-- 0617_02.xml 72 | | |-- MP_SEL_000201.jpg 73 | | |-- MP_SEL_000202.jpg 74 | | |-- MP_SEL_000203.jpg 75 | | |-- MP_SEL_000204.jpg 76 | | |-- ... 77 | |-- 0617_04 78 | | |-- 0617_03.xml 79 | | |-- MP_SEL_000601.jpg 80 | | |-- MP_SEL_000602.jpg 81 | | |-- MP_SEL_000603.jpg 82 | | |-- MP_SEL_000604.jpg 83 | | |-- ... 84 | ``` 85 | ## 2. 학습 86 | [Detectron2 AIHUB 데이터셋 튜토리얼(training, jupyter notebook)](https://github.com/visionNoob/detectron2_aihub_tutorial/blob/master/Detectron2_training_custom_dataset_(AIHub_sidewalk_datataset).ipynb) 87 | 88 | # Pretrained Models (13 classes): 89 | * Trained with 13 classes ([label.csv](https://www.dropbox.com/s/byecey0zebrn203/aihub_13_classes_label.csv?dl=0)) 90 | * You have to change your config ([example](https://www.dropbox.com/s/knvv6tyd6mna9ec/Base-RetinaNet.yaml)) 91 | -> Change Aspect Ratios of Anchor to 92 | ```[[0.65, 1.0, 2.47, 5.2, 18.12]]``` 93 | 94 | ### RetinaNet (trained with 13 claases) 95 | |Name|download| 96 | |-|-| 97 | |R50_3x|[model](https://www.dropbox.com/s/zuvj9qiuv5ntge9/retinanet_r_50_fpn_3x_aihub_final.pth) \| [metrics](https://github.com/visionNoob/detectron2_aihub_tutorial/blob/master/models/metrics.json) 98 | -------------------------------------------------------------------------------- /models/events.out.tfevents.1575349464.9bcafb0ea309.238.0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/visionNoob/detectron2_aihub_tutorial/652a162eed11f853e6c942a380f6fdf710232549/models/events.out.tfevents.1575349464.9bcafb0ea309.238.0 -------------------------------------------------------------------------------- /models/model.txt: -------------------------------------------------------------------------------- 1 | To download pertained model, 2 | Follow the command below : 3 | wget 'https://www.dropbox.com/s/zuvj9qiuv5ntge9/retinanet_r_50_fpn_3x_aihub_final.pth' --------------------------------------------------------------------------------