question about train FVC

zheng-yun commented 11 months ago

I want to train the model. Traceback (most recent call last): File "/mnt/4t/zy/FVC/main.py", line 304, in global_step = train(epoch, global_step) File "/mnt/4t/zy/FVC/main.py", line 202, in train seqout["rd_loss"].backward() File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py", line 396, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/init.py", line 173, in backward Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 253, in apply return user_fn(self, args) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 399, in wrapper outputs = fn(ctx, args) File "/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py", line 73, in backward deform_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.

WenBingo commented 11 months ago

想问下哥们你用什么卡训练的

WenBingo commented 11 months ago

I want to train the model. Traceback (most recent call last): File "/mnt/4t/zy/FVC/main.py", line 304, in global_step = train(epoch, global_step) File "/mnt/4t/zy/FVC/main.py", line 202, in train seqout["rd_loss"].backward() File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py", line 396, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/init.py", line 173, in backward Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 253, in apply return user_fn(self, args) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 399, in wrapper outputs = fn(ctx, args) File "/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py", line 73, in backward deform_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.

想问下 cd CreateI sh h265.sh $crf 1920 1024 这步骤是怎么执行的吗？

zheng-yun commented 11 months ago

先换到CreateI路径下，然后$crf替换成你要的QP就行

------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2023年10月11日(星期三) 晚上9:55 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

I want to train the model. Traceback (most recent call last): File "/mnt/4t/zy/FVC/main.py", line 304, in global_step = train(epoch, global_step) File "/mnt/4t/zy/FVC/main.py", line 202, in train seqout["rd_loss"].backward() File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py", line 396, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/init.py", line 173, in backward Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 253, in apply return user_fn(self, args) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 399, in wrapper outputs = fn(ctx, args) File "/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py", line 73, in backward deform_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.

想问下 cd CreateI sh h265.sh $crf 1920 1024 这步骤是怎么执行的吗？

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

WenBingo commented 11 months ago

感谢

------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2023年10月11日(星期三) 晚上9:57 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

先换到CreateI路径下，然后$crf替换成你要的QP就行

------------------ 原始邮件 ------------------ 发件人: @.>;
发送时间: 2023年10月11日(星期三) 晚上9:55 收件人: @.>;
抄送: @.>; @.>;
主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

I want to train the model. Traceback (most recent call last): File "/mnt/4t/zy/FVC/main.py", line 304, in global_step = train(epoch, global_step) File "/mnt/4t/zy/FVC/main.py", line 202, in train seqout["rd_loss"].backward() File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py", line 396, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/init.py", line 173, in backward Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 253, in apply return user_fn(self, args) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 399, in wrapper outputs = fn(ctx, args) File "/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py", line 73, in backward deform_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.

想问下 cd CreateI sh h265.sh $crf 1920 1024 这步骤是怎么执行的吗？

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.> — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

WenBingo commented 11 months ago

能在问下qt是什么吗？刚入视频方向...

------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2023年10月11日(星期三) 晚上9:57 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

先换到CreateI路径下，然后$crf替换成你要的QP就行

------------------ 原始邮件 ------------------ 发件人: @.>;
发送时间: 2023年10月11日(星期三) 晚上9:55 收件人: @.>;
抄送: @.>; @.>;
主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

I want to train the model. Traceback (most recent call last): File "/mnt/4t/zy/FVC/main.py", line 304, in global_step = train(epoch, global_step) File "/mnt/4t/zy/FVC/main.py", line 202, in train seqout["rd_loss"].backward() File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py", line 396, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/init.py", line 173, in backward Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 253, in apply return user_fn(self, args) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 399, in wrapper outputs = fn(ctx, args) File "/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py", line 73, in backward deform_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.

想问下 cd CreateI sh h265.sh $crf 1920 1024 这步骤是怎么执行的吗？

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.> — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

zheng-yun commented 11 months ago

你可以理解成一个参数，控制压缩的程度，QP越大，压缩的容量越小，视频质量越差，就是在量化的时候，如果QP是1就直接四舍五入取整，如果是2，要先除以2，再取整，越大，压缩的误差越大

------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2023年10月11日(星期三) 晚上10:04 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

能在问下qt是什么吗？刚入视频方向...

------------------ 原始邮件 ------------------ 发件人: @.>;
发送时间: 2023年10月11日(星期三) 晚上9:57 收件人: @.>;
抄送: @.>; @.>;
主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

先换到CreateI路径下，然后$crf替换成你要的QP就行

------------------&nbsp;原始邮件&nbsp;------------------
发件人: @.&gt;;
发送时间: 2023年10月11日(星期三) 晚上9:55
收件人: @.&gt;;
抄送: @.&gt;; @.&gt;;
主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

I want to train the model. Traceback (most recent call last): File "/mnt/4t/zy/FVC/main.py", line 304, in global_step = train(epoch, global_step) File "/mnt/4t/zy/FVC/main.py", line 202, in train seqout["rd_loss"].backward() File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py", line 396, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/init.py", line 173, in backward Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 253, in apply return user_fn(self, args) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 399, in wrapper outputs = fn(ctx, args) File "/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py", line 73, in backward deform_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.

想问下 cd CreateI sh h265.sh $crf 1920 1024 这步骤是怎么执行的吗？

—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID: @.&gt;
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

WenBingo commented 11 months ago

非常感谢

------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2023年10月11日(星期三) 晚上10:06 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

你可以理解成一个参数，控制压缩的程度，QP越大，压缩的容量越小，视频质量越差，就是在量化的时候，如果QP是1就直接四舍五入取整，如果是2，要先除以2，再取整，越大，压缩的误差越大

------------------ 原始邮件 ------------------ 发件人: @.>;
发送时间: 2023年10月11日(星期三) 晚上10:04 收件人: @.>;
抄送: @.>; @.>;
主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

能在问下qt是什么吗？刚入视频方向...

------------------&nbsp;原始邮件&nbsp;------------------
发件人: @.&gt;;
发送时间: 2023年10月11日(星期三) 晚上9:57
收件人: @.&gt;;
抄送: @.&gt;; @.&gt;;
主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

先换到CreateI路径下，然后$crf替换成你要的QP就行

------------------&amp;nbsp;原始邮件&amp;nbsp;------------------
发件人: @.&amp;gt;;
发送时间: 2023年10月11日(星期三) 晚上9:55
收件人: @.&amp;gt;;
抄送: @.&amp;gt;; @.&amp;gt;;
主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

I want to train the model. Traceback (most recent call last): File "/mnt/4t/zy/FVC/main.py", line 304, in global_step = train(epoch, global_step) File "/mnt/4t/zy/FVC/main.py", line 202, in train seqout["rd_loss"].backward() File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py", line 396, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/init.py", line 173, in backward Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 253, in apply return user_fn(self, args) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 399, in wrapper outputs = fn(ctx, args) File "/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py", line 73, in backward deform_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.

想问下 cd CreateI sh h265.sh $crf 1920 1024 这步骤是怎么执行的吗？

—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID: @.&amp;gt;
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: @.&gt;
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.> — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

WenBingo commented 11 months ago

您好，能再次请教一下Training data中的test.txt 是自己需要创建的空白text吗？我下载解压后的vimeo_septuplet中并没有test.txt文件

------------------ 原始邮件 ------------------ 发件人: "ZhihaoHu/PyTorchVideoCompression" @.>; 发送时间: 2023年10月11日(星期三) 晚上10:06 @.>; @.**@.>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

你可以理解成一个参数，控制压缩的程度，QP越大，压缩的容量越小，视频质量越差，就是在量化的时候，如果QP是1就直接四舍五入取整，如果是2，要先除以2，再取整，越大，压缩的误差越大

------------------ 原始邮件 ------------------ 发件人: @.>;
发送时间: 2023年10月11日(星期三) 晚上10:04 收件人: @.>;
抄送: @.>; @.>;
主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

能在问下qt是什么吗？刚入视频方向...

------------------&nbsp;原始邮件&nbsp;------------------
发件人: @.&gt;;
发送时间: 2023年10月11日(星期三) 晚上9:57
收件人: @.&gt;;
抄送: @.&gt;; @.&gt;;
主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

先换到CreateI路径下，然后$crf替换成你要的QP就行

------------------&amp;nbsp;原始邮件&amp;nbsp;------------------
发件人: @.&amp;gt;;
发送时间: 2023年10月11日(星期三) 晚上9:55
收件人: @.&amp;gt;;
抄送: @.&amp;gt;; @.&amp;gt;;
主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

I want to train the model. Traceback (most recent call last): File "/mnt/4t/zy/FVC/main.py", line 304, in global_step = train(epoch, global_step) File "/mnt/4t/zy/FVC/main.py", line 202, in train seqout["rd_loss"].backward() File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py", line 396, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/init.py", line 173, in backward Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 253, in apply return user_fn(self, args) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 399, in wrapper outputs = fn(ctx, args) File "/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py", line 73, in backward deform_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.

想问下 cd CreateI sh h265.sh $crf 1920 1024 这步骤是怎么执行的吗？

—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID: @.&amp;gt;
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: @.&gt;
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.> — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

WenBingo commented 11 months ago

能否可以加您的微信吗?我的微信号：18921762605

------------------ 原始邮件 ------------------ 发件人: "ZhihaoHu/PyTorchVideoCompression" @.>; 发送时间: 2023年10月11日(星期三) 晚上10:06 @.>; @.**@.>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

你可以理解成一个参数，控制压缩的程度，QP越大，压缩的容量越小，视频质量越差，就是在量化的时候，如果QP是1就直接四舍五入取整，如果是2，要先除以2，再取整，越大，压缩的误差越大

------------------ 原始邮件 ------------------ 发件人: @.>;
发送时间: 2023年10月11日(星期三) 晚上10:04 收件人: @.>;
抄送: @.>; @.>;
主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

能在问下qt是什么吗？刚入视频方向...

------------------&nbsp;原始邮件&nbsp;------------------
发件人: @.&gt;;
发送时间: 2023年10月11日(星期三) 晚上9:57
收件人: @.&gt;;
抄送: @.&gt;; @.&gt;;
主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

先换到CreateI路径下，然后$crf替换成你要的QP就行

------------------&amp;nbsp;原始邮件&amp;nbsp;------------------
发件人: @.&amp;gt;;
发送时间: 2023年10月11日(星期三) 晚上9:55
收件人: @.&amp;gt;;
抄送: @.&amp;gt;; @.&amp;gt;;
主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

I want to train the model. Traceback (most recent call last): File "/mnt/4t/zy/FVC/main.py", line 304, in global_step = train(epoch, global_step) File "/mnt/4t/zy/FVC/main.py", line 202, in train seqout["rd_loss"].backward() File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py", line 396, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/init.py", line 173, in backward Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 253, in apply return user_fn(self, args) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 399, in wrapper outputs = fn(ctx, args) File "/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py", line 73, in backward deform_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.

想问下 cd CreateI sh h265.sh $crf 1920 1024 这步骤是怎么执行的吗？

—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID: @.&amp;gt;
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: @.&gt;
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.> — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

WenBingo commented 11 months ago

你可以理解成一个参数，控制压缩的程度，QP越大，压缩的容量越小，视频质量越差，就是在量化的时候，如果QP是1就直接四舍五入取整，如果是2，要先除以2，再取整，越大，压缩的误差越大 … ------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2023年10月11日(星期三) 晚上10:04 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16) 能在问下qt是什么吗？刚入视频方向... ------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2023年10月11日(星期三) 晚上9:57 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16) 先换到CreateI路径下，然后$crf替换成你要的QP就行 ------------------&nbsp;原始邮件&nbsp;------------------ 发件人: @.&gt;; 发送时间: 2023年10月11日(星期三) 晚上9:55 收件人: @.&gt;; 抄送: @.&gt;; @.&gt;; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16) I want to train the model. Traceback (most recent call last): File "/mnt/4t/zy/FVC/main.py", line 304, in global_step = train(epoch, global_step) File "/mnt/4t/zy/FVC/main.py", line 202, in train seqout["rd_loss"].backward() File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py", line 396, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/init.py", line 173, in backward Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 253, in apply return user_fn(self, args) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 399, in wrapper outputs = fn(ctx, args) File "/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py", line 73, in backward deform_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead. 想问下 cd CreateI sh h265.sh $crf 1920 1024 这步骤是怎么执行的吗？ — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.&gt; — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

vimeo_septuplet中存在一个sep_testlist.txt文件，可能是数据集更新了。但在dataset.py中的def get_single_vimeo()中curnumber= int(y[-5:-4]) 得到的是“/”其中y = E:\FVC\data\dataset\vimeo_septuplet\sequences\00001\0626这种类型的数值。

想问下def get_single_vimeo()具体逻辑做什么样的数据处理？

WenBingo commented 11 months ago

你可以理解成一个参数，控制压缩的程度，QP越大，压缩的容量越小，视频质量越差，就是在量化的时候，如果QP是1就直接四舍五入取整，如果是2，要先除以2，再取整，越大，压缩的误差越大 … ------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2023年10月11日(星期三) 晚上10:04 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16) 能在问下qt是什么吗？刚入视频方向... ------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2023年10月11日(星期三) 晚上9:57 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16) 先换到CreateI路径下，然后$crf替换成你要的QP就行 ------------------&nbsp;原始邮件&nbsp;------------------ 发件人: @.&gt;; 发送时间: 2023年10月11日(星期三) 晚上9:55 收件人: @.&gt;; 抄送: @.&gt;; @.&gt;; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16) I want to train the model. Traceback (most recent call last): File "/mnt/4t/zy/FVC/main.py", line 304, in global_step = train(epoch, global_step) File "/mnt/4t/zy/FVC/main.py", line 202, in train seqout["rd_loss"].backward() File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py", line 396, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/init.py", line 173, in backward Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 253, in apply return user_fn(self, args) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 399, in wrapper outputs = fn(ctx, args) File "/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py", line 73, in backward deform_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead. 想问下 cd CreateI sh h265.sh $crf 1920 1024 这步骤是怎么执行的吗？ — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.&gt; — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

        y = os.path.join(rootdir, line.rstrip())
        curnumber = int(y[-5:-4])
        refnumber = curnumber - 2
        if refnumber < 1:
            continue

对这里面不是很明白想要做的处理

WenBingo commented 11 months ago

I want to train the model. Traceback (most recent call last): File "/mnt/4t/zy/FVC/main.py", line 304, in global_step = train(epoch, global_step) File "/mnt/4t/zy/FVC/main.py", line 202, in train seqout["rd_loss"].backward() File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py", line 396, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/init.py", line 173, in backward Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 253, in apply return user_fn(self, args) File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py", line 399, in wrapper outputs = fn(ctx, args) File "/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py", line 73, in backward deform_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.

想问下这个数据集是要做预处理吗？dataset.py对于vimeo_septuplet我没弄明白怎么处理的，vimeo_septuplet里面有多个文件夹，每个夹里面的图片也并不是都是同一个视频里面的？比如：vimeo_septuplet\sequences\00001\0001 这个文件夹里面图片就和vimeo_septuplet\sequences\00001\0005完全不一样

zheng-yun commented 10 months ago

我把可变性卷积里面的view全部改成了reshape

------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2023年10月30日(星期一) 下午2:29 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16)

我想训练模型。回溯（最近一次调用最后一次）：文件“/mnt/4t/zy/FVC/main.py”，第 304 行，在 global_step = train(epoch, global_step) 文件“/mnt/4t/zy/FVC/main.py” ”，第 202 行，火车 seqout["rd_loss"].backward() 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py”，第 396 行，在向后 torch.autograd.backward(self,gradient,retain_graph,create_graph,inputs=inputs) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/ init .py ", 第 173 行，向后 Variable._execution_engine.run_backward( # 调用 C++ 引擎来运行向后传递 File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd /function.py”，第 253 行，在应用中返回 user_fn(self, args) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py” ，第 399 行，在包装器输出中 = fn(ctx, args) 文件“/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py”，第 73 行，在向后的 fade_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not与输入张量的大小和步幅兼容（至少一个维度跨越两个连续的子空间）。使用 .reshape(...) 代替。

您好，我想请问您这个问题解决了吗？我尝试训练这个代码，遇到了同样的问题，可是代码中并没有view字样呀，很奇怪。

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

ShiLiangfan11 commented 10 months ago

我把可变性卷积里面的view全部改成了reshape … ------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2023年10月30日(星期一) 下午2:29 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16) 我想训练模型。回溯（最近一次调用最后一次）：文件“/mnt/4t/zy/FVC/main.py”，第 304 行，在 global_step = train(epoch, global_step) 文件“/mnt/4t/zy/FVC/main.py” ”，第 202 行，火车 seqout["rd_loss"].backward() 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py”，第 396 行，在向后 torch.autograd.backward(self,gradient,retain_graph,create_graph,inputs=inputs) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/ init .py ", 第 173 行，向后 Variable._execution_engine.run_backward( # 调用 C++ 引擎来运行向后传递 File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd /function.py”，第 253 行，在应用中返回 user_fn(self, args) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py” ，第 399 行，在包装器输出中 = fn(ctx, args) 文件“/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py”，第 73 行，在向后的 fade_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not与输入张量的大小和步幅兼容（至少一个维度跨越两个连续的子空间）。使用 .reshape(...) 代替。您好，我想请问您这个问题解决了吗？我尝试训练这个代码，遇到了同样的问题，可是代码中并没有view字样呀，很奇怪。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

我也没有找到view

LvMengfan233 commented 9 months ago

我把可变性卷积里面的view全部改成了reshape … ------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2023年10月30日(星期一) 下午2:29 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16) 我想训练模型。回溯（最近一次调用最后一次）：文件“/mnt/4t/zy/FVC/main.py”，第 304 行，在 global_step = train(epoch, global_step) 文件“/mnt/4t/zy/FVC/main.py” ”，第 202 行，火车 seqout["rd_loss"].backward() 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py”，第 396 行，在向后 torch.autograd.backward(self,gradient,retain_graph,create_graph,inputs=inputs) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/ init .py ", 第 173 行，向后 Variable._execution_engine.run_backward( # 调用 C++ 引擎来运行向后传递 File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd /function.py”，第 253 行，在应用中返回 user_fn(self, args) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py” ，第 399 行，在包装器输出中 = fn(ctx, args) 文件“/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py”，第 73 行，在向后的 fade_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not与输入张量的大小和步幅兼容（至少一个维度跨越两个连续的子空间）。使用 .reshape(...) 代替。您好，我想请问您这个问题解决了吗？我尝试训练这个代码，遇到了同样的问题，可是代码中并没有view字样呀，很奇怪。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

我把可变性卷积里面的view全部改成了reshape … ------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2023年10月30日(星期一) 下午2:29 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16) 我想训练模型。回溯（最近一次调用最后一次）：文件“/mnt/4t/zy/FVC/main.py”，第 304 行，在 global_step = train(epoch, global_step) 文件“/mnt/4t/zy/FVC/main.py” ”，第 202 行，火车 seqout["rd_loss"].backward() 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py”，第 396 行，在向后 torch.autograd.backward(self,gradient,retain_graph,create_graph,inputs=inputs) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/ init .py ", 第 173 行，向后 Variable._execution_engine.run_backward( # 调用 C++ 引擎来运行向后传递 File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd /function.py”，第 253 行，在应用中返回 user_fn(self, args) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py” ，第 399 行，在包装器输出中 = fn(ctx, args) 文件“/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py”，第 73 行，在向后的 fade_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not与输入张量的大小和步幅兼容（至少一个维度跨越两个连续的子空间）。使用 .reshape(...) 代替。您好，我想请问您这个问题解决了吗？我尝试训练这个代码，遇到了同样的问题，可是代码中并没有view字样呀，很奇怪。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

您好，我在训练fvc时也出现了deform_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead. 这一问题。看您也遇到了这样的问题，您解决的方式是把可变性卷积里面的view全部改成了reshape。想问下您改变的是哪个文件中的view？deform_conv.py文件中并未找到view，我尝试把deform_conv_cuda.cpp中的view改为reshape，还是会报这个错误。

bblgbr commented 9 months ago

我把可变性卷积里面的view全部改成了reshape … ------------------ 原始邮件 ------------------ 发件人: @.**>; 发送时间: 2023年10月30日(星期一) 下午2:29 收件人: @.**>; 抄送: @.**>; @.**>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16) 我想训练模型。回溯（最近一次调用最后一次）：文件“/mnt/4t/zy/FVC/main.py”，第 304 行，在 global_step = train(epoch, global_step) 文件“/mnt/4t/zy/FVC/main.py” ”，第 202 行，火车 seqout["rd_loss"].backward() 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py”，第 396 行，在向后 torch.autograd.backward(self,gradient,retain_graph,create_graph,inputs=inputs) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/ init .py ", 第 173 行，向后 Variable._execution_engine.run_backward( # 调用 C++ 引擎来运行向后传递 File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd /function.py”，第 253 行，在应用中返回 user_fn(self, _args) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py” ，第 399 行，在包装器输出中 = fn(ctx, _args) 文件“/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py”，第 73 行，在向后的 fade_conv_cuda.deform_conv_backward_parameterscuda( RuntimeError: view size is not与输入张量的大小和步幅兼容（至少一个维度跨越两个连续的子空间）。使用 .reshape(...) 代替。您好，我想请问您这个问题解决了吗？我尝试训练这个代码，遇到了同样的问题，可是代码中并没有view字样呀，很奇怪。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @_.__*>

我把可变性卷积里面的view全部改成了reshape … ------------------ 原始邮件 ------------------ 发件人: @.**>; 发送时间: 2023年10月30日(星期一) 下午2:29 收件人: @.**>; 抄送: @.**>; @.**>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16) 我想训练模型。回溯（最近一次调用最后一次）：文件“/mnt/4t/zy/FVC/main.py”，第 304 行，在 global_step = train(epoch, global_step) 文件“/mnt/4t/zy/FVC/main.py” ”，第 202 行，火车 seqout["rd_loss"].backward() 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py”，第 396 行，在向后 torch.autograd.backward(self,gradient,retain_graph,create_graph,inputs=inputs) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/ init .py ", 第 173 行，向后 Variable._execution_engine.run_backward( # 调用 C++ 引擎来运行向后传递 File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd /function.py”，第 253 行，在应用中返回 user_fn(self, _args) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py” ，第 399 行，在包装器输出中 = fn(ctx, _args) 文件“/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py”，第 73 行，在向后的 fade_conv_cuda.deform_conv_backward_parameterscuda( RuntimeError: view size is not与输入张量的大小和步幅兼容（至少一个维度跨越两个连续的子空间）。使用 .reshape(...) 代替。您好，我想请问您这个问题解决了吗？我尝试训练这个代码，遇到了同样的问题，可是代码中并没有view字样呀，很奇怪。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @_.__*>

您好，我在训练fvc时也出现了deform_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead. 这一问题。看您也遇到了这样的问题，您解决的方式是把可变性卷积里面的view全部改成了reshape。想问下您改变的是哪个文件中的view？deform_conv.py文件中并未找到view，我尝试把deform_conv_cuda.cpp中的view改为reshape，还是会报这个错误。

要改deform_conv_cuda.cpp中的deform_conv_backward_parameters_cuda函数，改完之后要重新编译安装，就好了

YY-RR-ZZ commented 9 months ago

我把可变性卷积里面的view全部改成了reshape … ------------------ 原始邮件 ------------------ 发件人: @.**>; 发送时间: 2023年10月30日(星期一) 下午2:29 收件人: @.**>; 抄送: @.**>; @.**>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16) 我想训练模型。回溯（最近一次调用最后一次）：文件“/mnt/4t/zy/FVC/main.py”，第 304 行，在 global_step = train(epoch, global_step) 文件“/mnt/4t/zy/FVC/main.py” ”，第 202 行，火车 seqout["rd_loss"].backward() 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py”，第 396 行，在向后 torch.autograd.backward(self,gradient,retain_graph,create_graph,inputs=inputs) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/ init .py ", 第 173 行，向后 Variable._execution_engine.run_backward( # 调用 C++ 引擎来运行向后传递 File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd /function.py”，第 253 行，在应用中返回 user_fn(self, _args) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py” ，第 399 行，在包装器输出中 = fn(ctx, _args) 文件“/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py”，第 73 行，在向后的 fade_conv_cuda.deform_conv_backward_parameterscuda( RuntimeError: view size is not与输入张量的大小和步幅兼容（至少一个维度跨越两个连续的子空间）。使用 .reshape(...) 代替。您好，我想请问您这个问题解决了吗？我尝试训练这个代码，遇到了同样的问题，可是代码中并没有view字样呀，很奇怪。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @_.__*>

我把可变性卷积里面的view全部改成了reshape … ------------------ 原始邮件 ------------------ 发件人: @.**>; 发送时间: 2023年10月30日(星期一) 下午2:29 收件人: @.**>; 抄送: @.**>; @.**>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16) 我想训练模型。回溯（最近一次调用最后一次）：文件“/mnt/4t/zy/FVC/main.py”，第 304 行，在 global_step = train(epoch, global_step) 文件“/mnt/4t/zy/FVC/main.py” ”，第 202 行，火车 seqout["rd_loss"].backward() 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py”，第 396 行，在向后 torch.autograd.backward(self,gradient,retain_graph,create_graph,inputs=inputs) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/ init .py ", 第 173 行，向后 Variable._execution_engine.run_backward( # 调用 C++ 引擎来运行向后传递 File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd /function.py”，第 253 行，在应用中返回 user_fn(self, _args) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py” ，第 399 行，在包装器输出中 = fn(ctx, _args) 文件“/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py”，第 73 行，在向后的 fade_conv_cuda.deform_conv_backward_parameterscuda( RuntimeError: view size is not与输入张量的大小和步幅兼容（至少一个维度跨越两个连续的子空间）。使用 .reshape(...) 代替。您好，我想请问您这个问题解决了吗？我尝试训练这个代码，遇到了同样的问题，可是代码中并没有view字样呀，很奇怪。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @_.__*>

您好，我在训练fvc时也出现了deform_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead. 这一问题。看您也遇到了这样的问题，您解决的方式是把可变性卷积里面的view全部改成了reshape。想问下您改变的是哪个文件中的view？deform_conv.py文件中并未找到view，我尝试把deform_conv_cuda.cpp中的view改为reshape，还是会报这个错误。

要改deform_conv_cuda.cpp中的deform_conv_backward_parameters_cuda函数，改完之后要重新编译安装，就好了

您好，请问你有复现成功FVC吗，我训练出来的结果好像差得多，不知道为啥

bblgbr commented 9 months ago

我把可变性卷积里面的view全部改成了reshape … ------------------ 原始邮件 ------------------ 发件人: @.**>; 发送时间: 2023年10月30日(星期一) 下午2:29 收件人: @.**>; 抄送: @.**>; @.**>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16) 我想训练模型。回溯（最近一次调用最后一次）：文件“/mnt/4t/zy/FVC/main.py”，第 304 行，在 global_step = train(epoch, global_step) 文件“/mnt/4t/zy/FVC/main.py” ”，第 202 行，火车 seqout["rd_loss"].backward() 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py”，第 396 行，在向后 torch.autograd.backward(self,gradient,retain_graph,create_graph,inputs=inputs) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/ init .py ", 第 173 行，向后 Variable._execution_engine.run_backward( # 调用 C++ 引擎来运行向后传递 File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd /function.py”，第 253 行，在应用中返回 user_fn(self, _args) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py” ，第 399 行，在包装器输出中 = fn(ctx, _args) 文件“/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py”，第 73 行，在向后的 fade_conv_cuda.deform_conv_backward_parameterscuda( RuntimeError: view size is not与输入张量的大小和步幅兼容（至少一个维度跨越两个连续的子空间）。使用 .reshape(...) 代替。您好，我想请问您这个问题解决了吗？我尝试训练这个代码，遇到了同样的问题，可是代码中并没有view字样呀，很奇怪。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @_.__*>

我把可变性卷积里面的view全部改成了reshape … ------------------ 原始邮件 ------------------ 发件人: @.**>; 发送时间: 2023年10月30日(星期一) 下午2:29 收件人: @.**>; 抄送: @.**>; @.**>; 主题: Re: [ZhihaoHu/PyTorchVideoCompression] question about train FVC (Issue #16) 我想训练模型。回溯（最近一次调用最后一次）：文件“/mnt/4t/zy/FVC/main.py”，第 304 行，在 global_step = train(epoch, global_step) 文件“/mnt/4t/zy/FVC/main.py” ”，第 202 行，火车 seqout["rd_loss"].backward() 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/_tensor.py”，第 396 行，在向后 torch.autograd.backward(self,gradient,retain_graph,create_graph,inputs=inputs) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/ init .py ", 第 173 行，向后 Variable._execution_engine.run_backward( # 调用 C++ 引擎来运行向后传递 File "/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd /function.py”，第 253 行，在应用中返回 user_fn(self, _args) 文件“/home/zy/anaconda3/envs/dvc/lib/python3.9/site-packages/torch/autograd/function.py” ，第 399 行，在包装器输出中 = fn(ctx, _args) 文件“/mnt/4t/zy/FVC/subnet/dcn/deform_conv.py”，第 73 行，在向后的 fade_conv_cuda.deform_conv_backward_parameterscuda( RuntimeError: view size is not与输入张量的大小和步幅兼容（至少一个维度跨越两个连续的子空间）。使用 .reshape(...) 代替。您好，我想请问您这个问题解决了吗？我尝试训练这个代码，遇到了同样的问题，可是代码中并没有view字样呀，很奇怪。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @_.__*>

您好，我在训练fvc时也出现了deform_conv_cuda.deform_conv_backward_parameters_cuda( RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead. 这一问题。看您也遇到了这样的问题，您解决的方式是把可变性卷积里面的view全部改成了reshape。想问下您改变的是哪个文件中的view？deform_conv.py文件中并未找到view，我尝试把deform_conv_cuda.cpp中的view改为reshape，还是会报这个错误。

要改deform_conv_cuda.cpp中的deform_conv_backward_parameters_cuda函数，改完之后要重新编译安装，就好了

您好，请问你有复现成功FVC吗，我训练出来的结果好像差得多，不知道为啥

正在复现，刚开始研究

ZhihaoHu / PyTorchVideoCompression

question about train FVC #16