antoine77340 / video_feature_extractor Goto Github PK

Is 3D ReNext101 feature similar to I3D feature ? Is it better than I3D？

In videocnn/README.md under the section "Requirements" the link http://johnvansickle.com/ffmpeg/releases/ffmpeg-release-64bit-static.tar.xz is dead. Instead I navigated to https://www.johnvansickle.com/ffmpeg/ and picked out the release that fit my system.

hey @antoine77340 , thanks for this work it's really intersting .!
i am trying to run the code , it's not really working for me and i guess that the problem with the csv file i provided .
i think i have some issues with the content of it .
can you provide an example for the csv file ?

thank you

Hi @antoine77340

Thank you so much for uploading this code its really working good and helpful for me.

I need to extract feature from video with time-based how can I do that could you help me to do this problem

problem statement :
if you give 3 videos as input I want to extraction frames based on time, how this way extracted feature from video

Thank you

Sorry, how can I convert my numpy element to classes? Example: "dog" "cat"Sorry, how can I convert my numpy element to classes? Example: "dog" "cat"

when i use pretrained 3D_resnext101 model, the dimensions of each frame is 2048. How can i get 1024 dimensions？

video_path, feature_path
video1.mp4, path_of_video1_features.npy
video2.webm, path_of_video2_features.npy

should be

video_path,feature_path
video1.mp4, path_of_video1_features.npy
video2.webm, path_of_video2_features.npy

otherwise we get a key error " feature_path"

hi, I am trying to run the feature extractor for my video pre-processing.

I want to generate the same dimensional feature output for all input videos of different lengths, is there an easy way to do this?

Thank you!

Dear antoine,

i have run 2 sample of mp4 files, it run without error. But i dont see where the extracted files are. It is not in the feature folder or anywhere else. Pls advice

folder feature is empty. CSV file:

video_path feature_path
/content/drive/MyDrive/video_sample/ABI - 3.mp4 /content/drive/MyDrive/video_feature/ABI - 3_features.npy
/content/drive/MyDrive/video_sample/ABI - 5.mp4 /content/drive/MyDrive/video_feature/ABI - 5_features.npy

.

Dear friend

I have to ask questions:

1. Is it possible to translate the features values and scores to the classes as output json?
2. Is it possible to catch these features at or with timestamps?

Please guide and help if this is possible.

Great to find your repo, @antoine77340!

I have 3 questions for you to answer:

1. The shape of my .npy file is (66, 2048), 66 is calculated as the total number of frames in the video divided by the batch size right?
2. The output shape I want is (10, 66, 2048), 10 is 10 crops argumentation, can you tell me how to do it?
3. If I want to change the resnet101 model to the resnet50 model, what components do I need to change in your repo?

Thank you so much!

ffprobe failed at: /home/machine21/Desktop/video_code/video_feature_extractor/video/input.mp4

Do I need to make the resizing of video to 112:112 ratio? or any other reason for this failure?

hello , thank you for yr amazing work
i keep getting these errors , if u can enlighten me how i can solve it

Traceback (most recent call last):
File "/path /to/extract.py", line 7, in
from model import get_model
File "/path/to/model.py", line 4, in
from videocnn.models import resnext
ModuleNotFoundError: No module named 'videocnn'

Can I ask you the 3d feature in your repo is c3d or i3d?
Thank you very much!!

.

Hi,
what is the shape of the data in npy file of every video?
I tried but it doesn't store in a proper format.
Can you guide me about it?

Thanks for the great work!
Regarding the question, I did try to change the frame rates within the files and it somehow worked with the 2D features, but for the 3D features, it still show result as per second. Can you show me which value I should change? Thanks!

Please advise on the feature extracted. I got (x,2048) and I pad them with the same length.
To run classification, do I need to change the dimension by adding channel? eg (x,2048,1)? or it doesn't necessary anymore.
(I'm still learning on this)

generate a csv of this form:

video_path,feature_path
absolute_path_video1.mp4,absolute_path_of_video1_features.npy absolute_path_video2.webm,absolute_path_of_video2_features.npy

Hello, how's the inputs look like? Are those patches from the frames of the same target? Or take one of the frames from different targets ?

Hi, great work there!
I noticed that the dimension of the output is not fixed and depends on video length(eg. some are 33x2048, some are 12x2048, etc)

What's the best way to get them to a single dim(eg. 1x2048)?

Thanks!

Based on https://github.com/kkroening/ffmpeg-python it seems that the correct way to install ffmpeg is:
pip install ffmpeg-python