search

yrcong / STTran

Spatial-Temporal Transformer for Dynamic Scene Graph Generation, ICCV2021
MIT License
188 stars 34 forks source link

Question of FastRCNN when using the sgdet mode #21

Closed Wh2018 closed 2 years ago

Wh2018 commented 2 years ago

Thanks for sharing the nice work!

I have successfully trained and tested the model in predcls mode, but when I want to train the model in the sgdet mode, I met a few problems.

In object_detector.py, fasterRCNN needs to return five values (Figure 1), but in faster_rnn.py, there are eight values to be retured (Figure 2), and there is no roi_features needed in object_detector.py

Could you tell me how to solve the problem requires, thank you very much 1 2

yrcong commented 2 years ago

Ye, we need return some features from the fasterrcnn. That's why the code is a little bit different.