airsplay/py-bottom-up-attention

Feature extraction demo (with attributes) not working ... ValueError: not enough values to unpack (expected 3, got 2)

ig-perez opened this issue · 2 comments

ig-perez commented

Hi,

Thanks for sharing your job. I'm interested in using this approach to extract features with attributes. I have installed Detectron's 2 v0.1.1 (around March 2020), PyTorch1.4 and I'm using faster_rcnn_R_101_C4_attr_caffemaxpool.yaml configuration file with http://nlp.cs.unc.edu/models/faster_rcnn_from_caffe_attr.pkl but I got above error when trying to run the demo.

I don't understand why predictor.model.roi_heads.box_predictor(feature_pooled) is returning a tuple of 2 tensors. It seems the pred_attr_logits are not returned.

Sorry if I'm missing some basic point. I'm new on the Computer Vision domain.

If interested please see related Colab Notebook here

Thanks!

skylie-xtj commented

Hi, have you solved it? I encountered the same question