search

yikuan8 / Transformers-VQA

An implementation that downstreams pre-trained V+L models to VQA tasks. Now support: VisualBERT, LXMERT, and UNITER
162 stars 28 forks source link

Extracting RoI Features #8

Open fds-uog opened 3 years ago

fds-uog commented 3 years ago

Hi,

Thanks for sharing your work. I was wondering how you extracted these visual features of the CXR images in the OpenI dataset using this code here.

I believe you had to modify some hyperparameters in the configuration file since, by using the same config used in their demos, Faster R-CNN doesn't seem to detect any bounding box. I tried to reduce the score_thresh=0 in the fast_rcnn_inference_single_image function and was able to see some bounding boxes detected. However, the features I extract don't match with yours.

I was wondering if you could explain a bit more in detail how to extract the same feature vectors and bounding box coordinates.

Thanks! Francesco

yikuan8 commented 3 years ago

You can also try this implementation. Honestly, I don't think it is possible or useful to extract the identical features.

petrusosk commented 2 years ago

Dear authors,

Would it be possible to share the code for the feature extraction? That would be very helpful.

Thank you!

whtitefall commented 1 year ago

Dear authors,

Would it be possible to share the code for the feature extraction? That would be very helpful.

Thank you!

Hi,

I'm wondering if you figure out how to extract features for customized dataset?