MagicSource

I have trained rfcn model and bbox accuracy got 97%, but when inference got no result, I am not convert model..... with: ``` BUILD=build/examples/FRCNN/demo_frcnn_api.bin $BUILD --gpu $gpu \ --model examples/FRCNN/rfcn-res50-voc/test_merged-atrous.proto...

``` Output Dir Is : examples/FRCNN/results/ I0813 11:08:07.946373 14934 demo_frcnn_api.cpp:70] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Demo for cat gray.jpg I0813 11:08:08.044909 14934 demo_frcnn_api.cpp:75] Predict cat gray.jpg cost 94.1493 ms. I0813 11:08:08.044935 14934 demo_frcnn_api.cpp:76] There...

@makefile Just simply using VOC, this data is veryfied training with zf-fasterrcnn, it worked. And the network loaded and trained, but inference got no result. Also, when I convert model...

@SuperPowerLF2 No idea so far.

It seems there were some nan out from first conv: ``` I0812 13:33:12.153046 24802 solver.cpp:231] Iteration 500 (2.45156 iter/s, 40.7904s/100 iters), loss = 7.33115 [ 500 / 140000 ] ->...

@makefile Thanks for your reply, same data I have trained on zf backbone, it's ok at least for 120000 iterations. From I can notice, it's still have BatchNorm layer, pretrained...

I am training only with VOC, what I mean is that, does rfcn-res50 or any other proto with resnets is OK to train? Have u tested with those configurations?

Just got loss nan on fasterrcnn with resnet50, with VOC data, not sure which reason for this.

For sure, start with train.sh: ``` #!/usr/bin/env sh # This script test four voc images using faster rcnn end-to-end trained model (ZF-Model) if [ ! -n "$1" ] ;then echo...

@345ishaan That would be very slow to generate every 3d box for every detected image patch.