YOLOv1 is new approach to object detection as a regression problem to spatially separated bounding boxes and associated class probabilities. YOLO is extremely fast, YOLO sees the entire image during training and test time so it implicitly encodes contextual information about classes as well as their appearance.
2023/07/05: Update DIoU, CIoU loss2023/07/03: Update GIoU loss2023/07/01: Update model weights Resnet18, Resnet34, Resnet50
| Backbone | Dataset | Training dataset | Valid dataset | Image size | mAP | mAP_50 | mAP_75 |
|---|---|---|---|---|---|---|---|
| Original paper | PASCAL VOC | trainval 2007+2012 | test2007 | 448x448 | -- | 0.63 | -- |
| ResNet18 + GIoU loss | PASCAL VOC | trainval 2007+2012 | test2007 | 448x448 | 0.41 | 0.64 | 0.44 |
| ResNet50 + GIoU loss | PASCAL VOC | trainval 2007+2012 | test2007 | 448x448 | 0.46 | 0.69 | 0.49 |
- Download Pascal VOC train+val 2012+2007
- Download Pascal VOC test 2007
Put all images, annotations, txt files in folder dataset/VOC folder as following:
├── VOC
├── images
├── trainval2007
├── 000005.jpg
├── 000007.jpg
├── trainval2012
├── test2007
├── images_id
├── trainval2007.txt
├── trainval2012.txt
├── test2007.txt
├── labels
├── trainval2007
├── 000005.xml
├── 000007.xml
├── trainval2012
├── trainval2007python -m src.train --model_type resnet18/resnet34/resnet50 --resume resume_most_recent_trainingpython -m src.eval --model_type resnet18/resnet34/resnet50 --weight_type path_to_weight_best.ptpython -m src.predict --model_type resnet18/resnet34/resnet50 --weight_path path_to_weight_best.pt --input_folder path_input_folder --output_folder path_to_output_folder --conf_thresh confidence_threshold --iou_thresh iou threshold