martkartasev / sepconv Goto Github PK

Could you provide the video "project/videos/see_you_again_540.mp4"?

And update the wiki with the findings

hello everymanito,
I would like to ask how to compile and start gpu acceleration, and what kind of environment is required?

best wishes

The the paper, they move the patch location in the first and third frame in opposite directions to enhance the motion, which is supposed to accelerate training.

Pyramid conversion needs to be fast enough

Possibly related to #9 and/or #12

Remember. We cant use numpy etc. We need to use torch variables, otherwise autograd wont work.

Usage of nn.Upsample causes the following warning:

UserWarning: Default upsampling behavior when mode=bilinear is changed to align_corners=False since 0.4.0. Please specify align_corners=True if the old behavior is desired. See the documentation of nn.Upsample for details.

The authors use SimpleFlow (https://docs.opencv.org/3.0-beta/modules/optflow/doc/dense_optflow.html) to estimate the optical flow in the images and discard it if it's below a certain threshold

File must be written with the proper syntax required by pip3, so that it's possible to execute the following: sudo pip3 install -r ./path/to/requirements.txt

Hi, I have tried to get the code running however I get this:
Traceback (most recent call last):
File "/usr/lib/python3.6/runpy.py", line 193, in _run_module_as_main
"main", mod_spec)
File "/usr/lib/python3.6/runpy.py", line 85, in _run_code
exec(code, run_globals)
File "/home/rf/sepconv/src/main.py", line 16, in
from src.model import Net
File "/home/rf/sepconv/src/model.py", line 10, in
from libs.sepconv.SeparableConvolution import SeparableConvolution
File "/home/rf/sepconv/libs/sepconv/SeparableConvolution.py", line 8, in
import libs.sepconv._ext.cunnex
File "/home/rf/sepconv/libs/sepconv/_ext/cunnex/init.py", line 3, in
from ._cunnex import lib as _lib, ffi as _ffi
ModuleNotFoundError: No module named 'libs.sepconv._ext.cunnex._cunnex'

I have tested this on 3 different systems with the required software version including cuda installed, searching up this issue has yielded no useful answers

Any suggestions as to what's going wrong?
Thanks in advance

Thanks for the great works!

I download the pretrained model as listed in the README.md recently.
Result is like this, https://imgur.com/Er3rBtf

I tried training on l1, l1+vgg, seems ok (at least not all noise)

I wonder if the current version model mismatch with the pretrained.pth provided?

all requirements are the same, except tensorboardx.

Thanks

src/SeparableConvolution_kernel.cu(66): error: pointer to incomplete class type is not allowed
...
src/SeparableConvolution_kernel.cu(170): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(170): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(171): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(171): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(173): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(173): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(174): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(174): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(176): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(176): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(177): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(177): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(179): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(179): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(180): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(180): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(181): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(181): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(182): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(182): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(189): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(189): error: pointer to incomplete class type is not allowed

src/SeparableConvolution_kernel.cu(190): error: pointer to incomplete class type is not allowed

In the article, the authors say they only use frames with enough texture for training (by visual inspection I assume, because they don't say they measure that)

When performing random crops in random images, accept them with a probability that depends on the estimated optical flow (the higher the flow, the more likely we are to use it as a training sample)

Not actually implemented by the authors

Detect frame changing to a completely different one

The conv in original is
==>Basic()
==>torch.nn.Conv2d(in_channels=intInput, out_channels=intOutput, ...)
Yours is
==>_conv_module()
==>torch.nn.Conv2d(in_channels, in_channels, kernel, stride, padding)
The channel is different.

Look into OpenCV python implementations of OF.

Hello,

Great work! I was just wondering if I can have a copy of the config file too. Maybe you have already uploaded it somewhere but I am missing it.

Thank you.