kaiseem/queryotr

[ECCV2022] Official PyTorch implementation of the paper "Outpainting by Queries"

queryotr's Issues

There was a problem executing your code. I wonder if you can help me

The following error occurred while executing the code.
please help me😥

Traceback (most recent call last):
File "main.py", line 103, in
g_grad_scale=g_grad_scaler)
File "/workspace/QueryOTR/engine.py", line 53, in train_one_epoch
D_losses.backward()
File "/opt/conda/envs/hgonet/lib/python3.7/site-packages/torch/_tensor.py", line 255, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs)
File "/opt/conda/envs/hgonet/lib/python3.7/site-packages/torch/autograd/init.py", line 149, in backward
allow_unreachable=True, accumulate_grad=True) # allow_unreachable flag
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [512]] is at version 1; expected version 0 instead. Hint: the backtrace further above shows the operation that failed to compute its gradient. The variable in question was changed in there or anywhere later. Good luck!

the link of BaiduNetdisk is unavailable

scenery Dataset, theBaiduNetDisk link is unavailable

实验结果与论文相差很大

谢谢前辈指点~

Hello, you are doing a great job. I have a few questions and I hope to get your answers.

1.Do you intend to open source your model ？
2.According to the data set you mentioned, how long has your model been trained, and on what equipment?
3.How well does this method work for larger images? Such as 1920 ✖ ️ 1080
4.What is the speed of reasoning? On what device ?
If you can answer, I will be very grateful! Thanks.

same question.

    same question.

Traceback (most recent call last):
File "D:\MyProject\PyThon\FlowInpainting\QueryOTR\main.py", line 100, in
train_one_epoch(opts, gen, cnn_dis, criterion, train_loader, opt_g, opt_d, torch.device('cuda'), epoch,
File "D:\MyProject\PyThon\FlowInpainting\QueryOTR\engine.py", line 52, in train_one_epoch
D_losses.backward()
File "D:\TOOLKITS\miniconda\lib\site-packages\torch_tensor.py", line 255, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs)
File "D:\TOOLKITS\miniconda\lib\site-packages\torch\autograd_init_.py", line 147, in backward
Variable._execution_engine.run_backward(
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [512]] is at version 1; expected version 0 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).

Originally posted by @MasterHow in #3 (comment)

Problem of reproduction.

Dear authors, thanks for your great work.
I have a little problem reproducing the results reported in the paper.
On the Scenery dataset, you mentioned the upper bound of IS score is 4.091, so did you only use the test set to compute the score or the whole dataset? If using test set only, how to select the test set?
In my experiments, when I use the total dataset to compute the IS, I get 3.744, and when I use the last 1000 images to compute the IS, I get 3.541. Could you give me some advice? Thanks a lot

结果网格化严重

你好！感谢你的分享！
我在尝试迁移你的工作到我的任务中，我想输入512x512的图片，然后让网络生成左边，输出512x1024的图片，只改了输入输出大小，但是结果比较差，对于复杂背景有网格现象，简单背景有的patch像素异常，如下图（左边原图，右边生成图，输入是原图的右半边），可能的原因是什么呢？是否需要改变patch的大小？

您好，复现结果与论文结果有差距

scenery6000数据集采用该论文的分法，epoch设置成300.也是在3090上训练的。
FID值是21.08， 论文里是20.366
IS是3.81，论文里是3.955
上面两个值与论文结果的差距是否是正常的？

此外，我也计算了PSNR值，这一项相差较大，我测出来是21.63，原论文是23.60.
训练代码没有进行任何修改

inference code

Nice work !,,,could you please share me your inference code?