Lei

Have u modified the file "train.py"? This file has a maximum line number 135, while your logo shows " File "train.py", line 149, in..."

Modifying decode_jpeg to decode_png has nothing to do with saving '.png' image, you should change the suffix of output name from '.jpg' to '.png'

large batch size or large input size?

> really slow. 20 seconds/step looks like your training runs on CPU only, try using nvidia-smi to see GPU status when training?

> Dataset and pre-trained model posted here: http://www.europe.naverlabs.com/Research/Computer-Vision/Learning-Visual-Representations/Deep-Image-Retrieval . Link was broken, authors fixed it recently. Thank you! I have downloaded the model and prototxt. I simply use the "model.caffemodel"...

> @beshining You may have to download and compile the caffe version of Faster-RCNN first. Thank you! I will try it. Hope the problem can be solved

> nope :( Have you fixed this problem now? It seems that we all have this problem :-)