I'm trying to reproduce the imagenet pretraining step, want to make sure that I'm using the same setting with you.
The solver I'm using comes from pva-fast-rcnn/models/pvanet/example_train/solver.prototxt
All images were resized into 256x256, and cropped to 192x192.
In pva-fast-rcnn/models/pvanet/imagenet/descriptions.txt, you mentioned multi-scale ILSVRC2012 training set. I'm curious about tools that you used when preparing the dataset. The tool that caffe provided seems not be able to put multi-scale images into the same lmdb file. It would be sweet if you could share your tools.
I'm trying to reproduce the imagenet pretraining step, want to make sure that I'm using the same setting with you.
In pva-fast-rcnn/models/pvanet/imagenet/descriptions.txt, you mentioned multi-scale ILSVRC2012 training set. I'm curious about tools that you used when preparing the dataset. The tool that caffe provided seems not be able to put multi-scale images into the same lmdb file. It would be sweet if you could share your tools.
Thanks