Contextdesc regional features extraction

[zhanghanduo]

Hi. For the extraction of regional features for the training of contextdesc, I don't know if I understand the steps:

1. Download the Retrieval model "reg.pb"
2. Resize input image to 448x448
3. Use the function "extract_regional_features" to get the corresponding reg_feat of the resized images to be trained: "reg_feat_list "
4. Get binary files

for reg_feat in reg_feat_list:
        reg_feat.astype(np.float32).tofile(save_path)

Is that correct?