Didn't analyze all the frames

Hello, everybody! My name is Esteban. This is the first time I comment in the forum and I hope you can help me with a little problem I have. I’ve started using deeplabcut software recently and I found the following:

A few days ago, thanks to the help of @Anna_Teruel, I made the first tracking of a video. We sent to the cluster 2 videos of 3 hours each, with 30 frames per second (a total of 10800 seconds, approximately 324000 frames and .avi format). Today when reviewing the .csv of coordinates I realized that I only have coordinates of 167000 frames approximately of the total 324000. This has happened in both videos and it seems that in both it has remained at 51%. What could be the problem?

Because this is the first time I’ve used the software I can’t detect exactly where the fault was, I don’t have much idea yet, sorry. Therefore, I leave in the message the .txt file (where all the processes of the cluster are gathered since the video is sent until the end of the analysis) in case you can help me.

Thank you everyone for your help!

deeplabcut_1931805.txt (6.2 MB)

very odd! Can you tell me a bit about your gpu and your frame size? the most important feature is not FPS (as DLC looks frame by frame), so I’m worried about some memory leak.

When it’s running can you check in the terminal the GPU usage by typing nvidia-smi and look at % GPU?

Hello @EMerinoM! We’ve introduced something new in 2.2b8 that lets you check the integrity of your video; this may take a little while on such a long video but that’d be cool if you could give it a try.

from deeplabcut.utils.auxfun_videos import VideoReader
video = VideoReader(path_to_your_video)
video.check_integrity()

This scans your video for errors and writes them to a log file in the same folder as your video. If it is not empty you’ll be warned, and you could upload it here :slight_smile:

3 Likes

Hello @MWMathis ! I’m very sorry for the delay in answering. To answer your question I had to ask the responsible of the cluster and he answered this morning. His answer was the next:

“Well, I don’t know exactly what they meant when they asked you about the GPU size. If it is to the RAM memory that each GPU has, it is 32 GB. In case it helps too, the model of the GPU is this: “Tesla V100-PCIE”. And the cluster has a limitation of use of one GPU by each user simultaneously”. (This is the literal translation of the mail you sent)

For the analysis of the videos we are using the computer cluster of the “Centro de investigación príncipe Felipe” which is located in Valencia, Spain.

As for the frame size, I don’t know very well what you mean, but I show you below the properties of my video:

  • Resolution 640x480
  • Total data/bit rate 192 kbps
  • 30 frames/second

I am very sorry for the delay in answering and thank you very much for the response :slight_smile:

I’m going to try it out this week.

Thank you very much for the answer :slight_smile: