Evaluation Results Are Very Low Only in --eval-only Mode (AP=0.0), But Normal During Training

[GuanyuChen8]

Hi, I’m training OneFormer for instance segmentation on a custom COCO-format dataset with two classes .

Everything works fine during training — the periodic evaluation shows reasonable AP values . However, when I run:

python train_net.py --config-file path/to/config.yaml --eval-only MODEL.WEIGHTS path/to/model_final.pth

I get extremely low (or 0.0) AP values across the board：

Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.000
...
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = -1.000