search

seominseok0429 / Real-world-Anomaly-Detection-in-Surveillance-Videos-pytorch

Real-world Anomaly Detection in Surveillance Videos- pytorch Re-implementation
114 stars 26 forks source link

i3d pretrained model #2

Closed marsplant closed 2 years ago

marsplant commented 3 years ago

thanks for your great work.

would you share your i3d pretrained model to extract the video clip feature ?

seominseok0429 commented 3 years ago

I'll share. Sorry for the late reply.

abhishekaich27 commented 3 years ago

@seominseok0429 it would be great if you could provide the model and script to extract i3D features.

t-v-k-git commented 2 years ago

please provide the scrip to extract I3d features

seominseok0429 commented 2 years ago

Ours full code available at CODE

coranholmes commented 2 years ago

Ours full code available at CODE

Hi, thank you for sharing the full codes. But it seems the feature extractor is still not included?

seominseok0429 commented 2 years ago

Oops. Model is missing. I'll upload it again

coranholmes commented 2 years ago

Oops. Model is missing. I'll upload it again

Please let me know if you have uploaded the latest version. Thank you so much!

seominseok0429 commented 2 years ago

@coranholmes This is the modified download link. https://drive.google.com/file/d/1xYsBiCSmXjE0BwoiH_Bcm4AWaA1pESHF/view?usp=sharing.

Thank!

coranholmes commented 2 years ago

@coranholmes This is the modified download link. https://drive.google.com/file/d/1xYsBiCSmXjE0BwoiH_Bcm4AWaA1pESHF/view?usp=sharing.

Thank!

Hi, thank you for the quick response. I have downloaded the new version, but it seems the same as the previous version. Would you please let me know which script is used to extract the features? Thank you.

zimengxueying commented 2 years ago

Have you solved the problem? I'm also confused about it. thx

@coranholmes This is the modified download link. https://drive.google.com/file/d/1xYsBiCSmXjE0BwoiH_Bcm4AWaA1pESHF/view?usp=sharing. Thank!

Hi, thank you for the quick response. I have downloaded the new version, but it seems the same as the previous version. Would you please let me know which script is used to extract the features? Thank you.

coranholmes commented 2 years ago

Have you solved the problem? I'm also confused about it. thx

@coranholmes This is the modified download link. https://drive.google.com/file/d/1xYsBiCSmXjE0BwoiH_Bcm4AWaA1pESHF/view?usp=sharing. Thank!

Hi, thank you for the quick response. I have downloaded the new version, but it seems the same as the previous version. Would you please let me know which script is used to extract the features? Thank you.

Well no... But you can consider trying this repo if you just want the i3d extractor.

zimengxueying commented 2 years ago

Have you solved the problem? I'm also confused about it. thx

@coranholmes This is the modified download link. https://drive.google.com/file/d/1xYsBiCSmXjE0BwoiH_Bcm4AWaA1pESHF/view?usp=sharing. Thank!

Hi, thank you for the quick response. I have downloaded the new version, but it seems the same as the previous version. Would you please let me know which script is used to extract the features? Thank you.

Well no... But you can consider trying this repo if you just want the i3d extractor.

Thanks a lot! I will try it!

zimengxueying commented 2 years ago

Have you solved the problem? I'm also confused about it. thx

@coranholmes This is the modified download link. https://drive.google.com/file/d/1xYsBiCSmXjE0BwoiH_Bcm4AWaA1pESHF/view?usp=sharing. Thank!

Hi, thank you for the quick response. I have downloaded the new version, but it seems the same as the previous version. Would you please let me know which script is used to extract the features? Thank you.

Well no... But you can consider trying this repo if you just want the i3d extractor.

Another question, the original paper used i3d two-stream, but the link is resnet, is the speed and accuracy a little different right?

coranholmes commented 2 years ago

Have you solved the problem? I'm also confused about it. thx

@coranholmes This is the modified download link. https://drive.google.com/file/d/1xYsBiCSmXjE0BwoiH_Bcm4AWaA1pESHF/view?usp=sharing. Thank!

Hi, thank you for the quick response. I have downloaded the new version, but it seems the same as the previous version. Would you please let me know which script is used to extract the features? Thank you.

Well no... But you can consider trying this repo if you just want the i3d extractor.

Another question, the original paper used i3d two-stream, but the link is resnet, is the speed and accuracy a little different right?

Which paper are you referring to? For “Real world Anomaly Detection in Surveillance Videos”, I am not very sure. For “Weakly-supervised Video Anomaly Detection with Robust Temporal Feature Magnitude Learning”, they mention somewhere they are using the resnet as the backbone for I3D feature extraction. There may be some difference but won't be so significant I suppose.

zimengxueying commented 2 years ago

two stream that is this "Real-world-Anomaly-Detection-in-Surveillance-Videos-pytorch",the other paper you mentioned I have also read. I'm a little confused that the repo you provide, the dimension of the extracted feature is (n/16,2048) right? n is the length of one video, however, this repo provided (32,1024)for rgb and(32,1024)for optical flow."32" is fixed by divided the video by 32 segments even if the length of video is different. But n/16 is not fixed, so that is my question, or I’m misunderstanding, could you correct me? Sorry to bother you.

coranholmes commented 2 years ago

two stream that is this "Real-world-Anomaly-Detection-in-Surveillance-Videos-pytorch",the other paper you mentioned I have also read. I'm a little confused that the repo you provide, the dimension of the extracted feature is (n/16,2048) right? n is the length of one video, however, this repo provided (32,1024)for rgb and(32,1024)for optical flow."32" is fixed by divided the video by 32 segments even if the length of video is different. But n/16 is not fixed, so that is my question, or I’m misunderstanding, could you correct me? Sorry to bother you.

I only read RTFM source codes in details. They have some preprocessing in training phase, so the n/16 segments would be finally transformed to 32. see here

zimengxueying commented 2 years ago

two stream that is this "Real-world-Anomaly-Detection-in-Surveillance-Videos-pytorch",the other paper you mentioned I have also read. I'm a little confused that the repo you provide, the dimension of the extracted feature is (n/16,2048) right? n is the length of one video, however, this repo provided (32,1024)for rgb and(32,1024)for optical flow."32" is fixed by divided the video by 32 segments even if the length of video is different. But n/16 is not fixed, so that is my question, or I’m misunderstanding, could you correct me? Sorry to bother you.

I only read RTFM source codes in details. They have some preprocessing in training phase, so the n/16 segments would be finally transformed to 32. see here

Okay, let me read it. Thx for your patience!