feature extraction with docker: RuntimeWarning: invalid value encountered in greater_equal

Hi,

I was able to run the feature extraction script, `scripts/ extract_imgfeat.sh` with the Docker image from https://hub.docker.com/r/chenrocks/butd-caffe/tags?page=1&ordering=last_updated

It processed ~750 images before I encountered this:

`/src/tools/../lib/rpn/proposal_layer.py:184: RuntimeWarning: invalid value encountered in greater_equal`
`keep = np.where((ws >= min_size) & (hs >= min_size))[0] `
`F0607 05:45:37.953456 123 roi_pooling_layer.cu:87] Check failed: error == cudaSuccess (9 vs. 0) invalid configuration argument`

Do you have any pointers on what I could possibly try to fix this? Thank you!

Start by reproducing the failure with scripts/ extract_imgfeat.sh in the referenced Docker image. Inspect proposal_layer.py:184 and roi_pooling_layer.cu:87 around the reported warning and CUDA error. Done means feature extraction processes the image set without the runtime warning or invalid-configuration crash.

Written by the indexing model from the issue text.