Yann Henon

I'm not sure I fully understand what kind of example is being sought - I do have a repo here https://github.com/yhenon/keras-frcnn that show bounding box prediction using faster-rcnn. But it's...

Your best bet is to crop your images into smaller segments, so it does not need to be resized as aggressively.

Note that this repo is somewhat old and I haven't maintained it too well, so I cannot guarantee your results.

This is a good idea. However, you didn't apply the regression to the anchors quite right in your code. I can't train a network right now, the code below should...

Tell me how it goes, I might be interested in merging this

Interesting, it might be necessary to change the relative weighting of the regression vs classification. Let me look at that

I don't really have time to work on this project, but I believe the gap in performance is due to the limited data augmentation in this implementation (only flipping). It...

The important part is here: `ValueError: invalid literal for int() with base 10: '245.0' ` It looks like your class file is messed up somehow? Could you post it here?

- You need to have only 1 bounding box per line, i.e.: /data/imgs/img_001.jpg,837,346,981,456,wheat-head /data/imgs/img_002.jpg,215,312,279,391,wheat-head /data/imgs/img_002.jpg,22,5,89,84,wheat-head Values should be int, not float, so `942.0` -> `942` On Sat, Jul 18, 2020...

It's kind of confusing because of the paper text not fully explaining it. The smooth loss transitions from l1 to l2 behaviour at a point, See here for an explanation:...