Hi, thanks for your nice work. I am trying to replicate the results from the train_brightening procedure, however, I am unable to recreate the visual results of the enhanced data that you are providing. My enhanced saved images are much more noisy.

Hi, thanks for your great work! I want to know how you split DarkFace in to train&val set. Specifically, which 500 images are selected as val images? Moreover, are models in Table. 1 in the paper evaluated on val set or test set? Looking forward to your reply, 3Q!

Hello, amazing work, thanks for sharing the code, I have one question

It seems the test.py just save the detection results in a file, can you please tell me how to mAP value?

Hello, when I run "python train.py --multigpu" with 4 GPUs in pytorch1.7, I meet the following error, Traceback (most recent call last): File "train.py", line 403, in loss.backward() File "/root/anaconda3/lib/python3.7/site-packages/torch/tensor.py", line 221, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph) File "/root/anaconda3/lib/python3.7/site-packages/torch/autograd/init.py", line 132, in backward allow_unreachable=True) # allow_unreachable flag RuntimeError: cuDNN error: CUDNN_STATUS_BAD_PARAM You can try to repro this exception using the following code snippet. If that doesn't trigger the error, please include your original repro script when reporting this issue.

import torch torch.backends.cuda.matmul.allow_tf32 = True torch.backends.cudnn.benchmark = False torch.backends.cudnn.deterministic = False torch.backends.cudnn.allow_tf32 = True data = torch.randn([2, 64, 4, 4], dtype=torch.float, device='cuda', requires_grad=True) net = torch.nn.Conv2d(64, 64, kernel_size=[3, 3], padding=[1, 1], stride=[1, 1], dilation=[1, 1], groups=1) net = net.cuda().float() out = net(data) out.backward(torch.randn_like(out)) torch.cuda.synchronize() ConvolutionParams data_type = CUDNN_DATA_FLOAT padding = [1, 1, 0] stride = [1, 1, 0] dilation = [1, 1, 0] groups = 1 deterministic = false allow_tf32 = true input: TensorDescriptor 0x7f2944305e20 type = CUDNN_DATA_FLOAT nbDims = 4 dimA = 2, 64, 4, 4, strideA = 1024, 16, 4, 1, output: TensorDescriptor 0x7f2944321900 type = CUDNN_DATA_FLOAT nbDims = 4 dimA = 2, 64, 4, 4, strideA = 1024, 16, 4, 1, weight: FilterDescriptor 0x7f29442f7740 type = CUDNN_DATA_FLOAT tensor_format = CUDNN_TENSOR_NCHW nbDims = 4 dimA = 64, 64, 3, 3, Pointer addresses: input: 0x7f267b5fd400 output: 0x7f267d77ba00 weight: 0x7f27d7b2f600 Additional pointer addresses: grad_output: 0x7f267d77ba00 grad_weight: 0x7f27d7b2f600 Backward filter algorithm: 3

Have you ever encountered this bug?

I want to test the effect on CodaLab, but prompt Submission upload has had been disabled. See the new instance at: https://codalab.lisn.upsaclay.fr/. I can't find the same item in the new site, how do I test it? Thank you

You want to try to reproduce your code, but you can't find the ExDark dataset corresponding to your exdark_test.txt. You can package and send me a complete exdark dataset and .txt. Type of tag file? Email:[email protected], thank you very much.

During the pre-training of the classification head and the comparison learning head, as well as during the Joint High-Low Adaptation training, the loss does not converge and oscillates repeatedly. Can you provide me with your gradient descent curve?

Have you met this problem?

self.queue[ind, :, ptr:ptr + batch_size] = keys.T RuntimeError: The expanded size of the tensor (6) must match the existing size (16) at non-singleton dimension 1. Target sizes: [128, 6]. Tensor sizes: [128, 16]