The master branch works with PyTorch 1.1 or higher.
mmdetection-plus is a enhanced object detection codebase for my research and projects based on mmdetection, which is developed by Multimedia Laboratory, CUHK.
- Modify mmdet/apis/inference.py
- In line 143:
show=out_file is None, out_file=out_file
- In line 143:
- Add inference code for easily building detection web service
- Add VOC-COCO, YOLO-VOC data converter
- Install mmdetection.
- Prepare data.
- Modify config py:
bbox_head: num_classes = cls_num + 1
total_epoch
- Modify mmdet/core/evaluation/class_names.py
- Modify mmdet/datasets/voc.py:
CLASSES = ('xx', 'xxx')
Note: If you only have 1 class, use CLASSES = [‘xx’], or you will get some trouble.
- Re-install:
python3 setup.py develop
The detailed description of training/test procedure can be found in GETTING_STARTED.