Why are the AP scores all 0 when running test.py?

[Connor Lee]

• I'm using the pretrained kitti weights.
• I'm testing on the KITTI 2d dataset downloaded from the official site.
• I converted the kitti labels to coco format and put them under PyTorch-YOLOv3-kitti\data\kitti\labels\test
• Test images are under PyTorch-YOLOv3-kitti\data\kitti\image\test

[Deng An]

make sure val.txt have test image path. cause I set test_path = data_config["valid"] in test.py, you can change it by youself.
it is my results:
Detecting objects: 100%|████████████████████████| 47/47 [01:01<00:00, 1.24s/it]
Computing AP for class '0': 100%|███████████| 747/747 [00:00<00:00, 7063.08it/s]
Computing AP for class '1': 100%|██████████| 747/747 [00:00<00:00, 73969.95it/s]
Computing AP for class '2': 100%|█████████| 747/747 [00:00<00:00, 153887.28it/s]
Computing AP for class '3': 100%|██████████| 747/747 [00:00<00:00, 47469.02it/s]
Computing AP for class '4': 100%|█████████| 747/747 [00:00<00:00, 405060.77it/s]
Computing AP for class '5': 100%|██████████| 747/747 [00:00<00:00, 86269.76it/s]
Computing AP for class '6': 100%|█████████| 747/747 [00:00<00:00, 263821.58it/s]
Computing AP for class '7': 100%|█████████| 747/747 [00:00<00:00, 153736.27it/s]
Average Precisions:

• Class '0' - AP: 0.9037875496395236
• Class '1' - AP: 0.9242369542262239
• Class '2' - AP: 0.9158339295047374
• Class '3' - AP: 0.7671271411530298
• Class '4' - AP: 0.7692063492063491
• Class '5' - AP: 0.783446162852444
• Class '6' - AP: 0.8988244420575247
• Class '7' - AP: 0.6701750837797066
  mAP: 0.8290797015524424

[Connor Lee]

Ahh thats right, thanks!

[DHPark]

Excuse me, I also Got 0 AP .
I put kitti 1000 images to kitti/image/test and 1000 transformed labels to kitti/label/test.

I set val.txt point images in kitti/image/test

What am I doing wrong??

[Connor Lee]

The path in dataset.py points to 'images' rather than 'image'. Change one or the other to get it to work.

[DHPark]

@connorlee77 Oh I see. Thanks a lot.