AIWintermuteAI
commented
3 years ago Thanks for filling the from. From what I see in your screenshots, you can get 0.1 mAP, which is not zero, but very low.
Judging by this line 5/5 [==============================] - 1s 141ms/step it only takes 1 second for full pass over dataset - so either you have a very powerful machine or your dataset is extremely small. Remember that at very least 250-500 instances is required per class to achieve satisfactory results, if you use transfer learning. Much more is needed if you train from scratch.
This is all I can say without looking at a)config b)dataset
wwzh2015
Describe the bug Multi Object Detect Is Wrong
To Reproduce Steps to reproduce the behavior: Train multi classes object detector by pascal_20_detector.json.
Expected behavior If I just have only one class object that‘s mAp calculate is true, However when multi class objects(class>1) that's mAp is 0 or very low! and object detect have nothing result. what's happened?
Screenshots https://ftp.bmp.ovh/imgs/2021/07/9a67e2599fcda7c9.png https://ftp.bmp.ovh/imgs/2021/07/a9db47136bcfbd7d.jpg
Environment (please complete the following information):
Additional context Why the four classes ap also is 0? battery 0.0000 cardboard 0.0000 glass 0.0000 metal 0.0000 mAP: 0.0000 mAP did not improve from 0. Epoch 42/100 5/5 [==============================] - 1s 141ms/step - loss: 1.0767 - val_loss: 0.7849