DmitryUlyanov ( Dmitry Ulyanov )

Hi, the issue is relevant but I don't have time to check YUV now. Thanks for the clarification, I will look for a workaround on the color normalization.

Best, Dmitry

On Wed, Jun 17, 2020 at 11:26 AM Roman Arzumanyan [email protected] wrote:

Hi @DmitryUlyanov https://github.com/DmitryUlyanov Is this issue still relevant?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/NVIDIA/VideoProcessingFramework/issues/63#issuecomment-645232036, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABTKOZFULCNHVPETI76YXALRXB43JANCNFSM4N5KF2JA .

DmitryUlyanov

comment created time in 2 months

issue openedNVIDIA/VideoProcessingFramework

RGB frames extracted with VPF are different to FFmpeg

Describe the bug Hi, I want to process the video frame by frame using pytorch and I followed the example on wiki. Previously I used ffmpeg to extract frames and load them sequentially. I wanted to validate if VPF decodes frames in the same way but it turns out that there is some discrepancy in the appearance.

Here is the frame extracted with VPF: 00000

And with FFMpeg 00001

Please find here the video. And here is the exact code I am running.

Expected behavior The images should be the same

Desktop (please complete the following information):

OS: Ubuntu 18
Nvidia driver version 440.82
CUDA Version 10.2
Video Codec SDK Version 9.1.23
Python Version 3.8

created time in 2 months

push eventhse-aml/competitive-data-science

Dmitry Ulyanov

commit sha f8f63a749fb1515239c2ad1b7146951f9588344a

view details

Dmitry Ulyanov

commit sha b43fc8ae2109837c7bb0acb08fc8bebd94c6e269

Merge branch 'master' of https://github.com/hse-aml/competitive-data-science

view details

Dmitry Ulyanov

commit sha 3d2dc15420260c02f383c41011feecdb88dbe616

Merge branch 'master' of https://github.com/hse-aml/competitive-data-science

view details

push time in 2 months

MemberEvent

push eventhse-aml/competitive-data-science

Vishal Sharma

commit sha 1be0e9e5a7b5a33ad769a38bc0e6366dcf999707

fixing typo

view details

Dmitry Ulyanov

commit sha 68728a4127712f4991bd2454582d0918634758bf

Merge pull request #16 from vishalshar/patch-1 fixing typo

view details

push time in 2 months

PR merged hse-aml/competitive-data-science

fixing typo

+1 -1

1 commit in vishalshar/competitive-data-science/tree/patch-1 hse-aml/competitive-data-science/tree/master

0 comment

1 changed file

vishalshar

pr closed time in 2 months

PR opened cnr-isti-vclab/meshlab

Fix openMP related crash

Fixes https://github.com/cnr-isti-vclab/meshlab/issues/716

+1 -1

1 commit in DmitryUlyanov/meshlab/tree/patch-3 cnr-isti-vclab/meshlab/tree/master

0 comment

1 changed file

pr created time in 2 months

push eventDmitryUlyanov/meshlab

Dmitry Ulyanov

commit sha 1116b251617bc87ed79624b6dd54515895055748

Update shared.pri fixes https://github.com/cnr-isti-vclab/meshlab/issues/716

view details

push time in 2 months

issue commentcnr-isti-vclab/meshlab

undefined symbol: GOMP_parallel

Should -fopenmp be added here https://github.com/cnr-isti-vclab/meshlab/blob/master/src/shared.pri#L13 ?

DmitryUlyanov

comment created time in 2 months

issue openedcnr-isti-vclab/meshlab

undefined symbol: GOMP_parallel

Check all platforms you experienced the issue (change to `[x]`)

[ ] Windows
[ ] Mac OS
[x ] Linux
[ ] Every platform

What type of issue is?

[ ] bug or unexpected behavior
[x] crash
[ ] feature request
[ ] other (specify...)

Describe your issue

I get error message

/home/dmitry/meshlab_portable/meshlab: symbol lookup error: /home/dmitry/meshlab_portable/plugins/libfilter_select.so: undefined symbol: GOMP_parallel

Whenever I click "select outliers" filter. The problem is in openMP linking. I previously submitted a PR for another filter: https://github.com/cnr-isti-vclab/meshlab/pull/649 but I don't know where to set the flag for this filter. There should be a one-line fix this issue.

created time in 2 months

issue openedpytorch/pytorch

Make torch.cross `dim` parameter work intuitively

Hi PyTorch team,

Thanks a lot for maintaining such a great library.

Problem

torch.cross has dim arg, and by default dim=-1. However, it turns out, that if dim is not set, that dim is chosen to be the first dimension of size 3. This is very weird for me as a user and I spent half a day looking for a bug because I did not read the doc in full.

Solution

Either to set default dim to None pushing the reader to go further in docs to understand what it means. A better way would be to remove "first dimension of size 3" thing and just use dim parameter.

created time in 3 months

issue commentNVIDIA/DALI

Enable VideoReader to read variable frame-rate videos via flag

HI @a-sansanwal , any progress on this one?

lucadealfaro