Class-wise mAP only showing results for a few classes during evaluation

[gauricollab09]

问题确认 Search before asking

• 我已经搜索过问题，但是没有找到解答。I have searched the question and found no related answer.

请提出你的问题 Please ask your question

Hi all,

I'm encountering an issue while evaluating my object detection model that I trained using the --classwise argument in the command as follows:
python3 tools/eval.py -c configs/ppyoloe/voc/ppyoloe_plus_crn_s_30e_voc_NC.yml --classwise --amp

The evaluation only displays the AP for the first two classes in the label_list.txt, but I have multiple classes in my dataset. Here are the details of my setup:

Environment:

PaddleDetection version: 3.0.0-betal
PaddlePaddle version: release/2.5
OS: Ubuntu 20.04

What I have tried so far:

I verified that my dataset contains annotations for all 19 classes.
I used a lower confidence threshold (--conf_thres 0.001) to include more detections, but the issue persists.
I checked the label_list.txt.

Could someone else me with this? Thank you!

[liu-jiaxuan]

Hi, please check if other output results (e.g., PR curve) include all categories

[gauricollab09]

@liu-jiaxuan It contains the curves for only the two categories - cup and person.

[liu-jiaxuan]

Hi, the PaddleDetection beta version may have potential problems. Please use PaddleX first. You can refer to the tutorial