greatwallet / pytorch-implemented-deep-sr-itm Goto Github PK

Hey sorry to bother you.

How can I input png 8bit and png 16bit but in RGB?

Or is it possible, if I convert them to YUV, to read them from .hdr float32 and copare them to 8bit png?

my script is here

Hello,

MSU Graphics & Media Lab Video Group has recently launched HDR Video Reconstruction Benchmark.

If you are interested in participating, you can add your algorithm following the submission steps: Submit for MSU HDR Video Reconstruction Benchmark

We would be grateful for your feedback on our work!

Could you provide the pretrained model for just having a try? Thanks.

I try to run python test.py under env you mentioned. unlucky, the png is black.
model_path=./checkpoint/full_net/full_net_270.pth
2 graphic hardwares : Quadro TRX 2000
only changed code is: validate=false.

Hi
Since we can konw from the original implement that the author used ffmpeg to translate mp4 to yuv, but the train_dataset is in yuv format and packed into .mat file, how can we translate the single yuv file to [png] as the first step you proposed ?

Thanks!

网络的构建是对的，但由于命名不同，导致训练好的base_net参数加载到full_net时会有些问题。
例如，在base_net里，由于没有modulation，所以命名全部都是rb_base，而到了full_net，由于modulation的加入，个别模块的命名变为了rmb_base，但实际上模块在网络中的位置是相同的，base_net中的第2,4,6个rb_base模块的参数应该载入到当前full_net中的第1,2,3个rmb_base模块中，由于命名不同的原因，实际上应该是无法加载的。同时，当前full_net中的第2,3个rb_base应该对应base_net的第3,5个rb_base，由于命名的原因，应该也没有正确载入。
同理，另一支路也有相同的问题。