xian-bei / clut Goto Github PK

will cuda9.0 works?

Do you know how to convert raw data to xyz colorspace

Hi, thanks for this novel works. I am trying to training CLUT-Net on my computer with FiveK datasets. When starting training, I find the tvmn loss is nearly equal to zero(about 1e-8). Dose it normal, or Is there something wrong with my training config?

Could you please provide the visulization code for grid occupancy? I didn't find them in the /utils files.

请问为什么CLUTNet中的forward函数返回的是img_res + img_org呢？

I've reproduced the experiment with the setting of ["CLUTNet", "20+05+20"] and ['CLUTNet', '20+05+10'] in parameters.py. The dataset is fiveK and the PSNR is 25.22dB and 25.18dB for the best. Why the results are different with the data in the paper?

When I ran the test command: python test.py --model HashLUT 6+13 SmallBackbone --epoch 367, an error occurred. The error message is as follows:
6 tables of range:9-64, T:13, b:1.480 compensate: True SmallBackbone backbone Traceback (most recent call last): File "test.py", line 60, in <module> model.load_state_dict(ckpt, strict=True) File "site-packages/torch/nn/modules/module.py", line 1224, in load_state_dict self.__class__.__name__, "\n\t".join(error_msgs))) RuntimeError: Error(s) in loading state_dict for HashLUT: size mismatch for hashluts.params: copying a param with shape torch.Size([73392]) from checkpoint, the shape in current model is torch.Size([73648]).

Please help me to get the retouched photos on the ppr10k dataset. Thanks!