search

chenyuntc / simple-faster-rcnn-pytorch

A simplified implemention of Faster R-CNN that replicate performance from origin paper
Other
4k stars 1.14k forks source link

extracting features for specific boxes #24

Open volkancirik opened 6 years ago

volkancirik commented 6 years ago

Hello,

first of all thanks for your great efforts!

Is it possible to extract features for specific boxes? or even fine-tune the f-rcnn based on generated those features. An example case would be let's say you have annotated boxes for visual qa and would like to use f-rcnn as feature extractor and even finetune the pretrained model. any tips towards this direction?

chenyuntc commented 6 years ago

I Don't know whether I get your idea, but I think you'll need to custom the ROIHead.

If you don't rely on faster rcnn to find boxes -- I mean you only need pretrained models to extract features for specified boxes, then you don't need faster rcnn.

volkancirik commented 6 years ago

So you are suggesting using a standard pre-trained model (VGG,Resnet) as feature extractor for bounding boxes instead of f-rcnn?

chenyuntc commented 6 years ago

Yes!