Could I get original TSN features?
Pilhyeon opened this issue · comments
Hello,
I'm trying to experiment with this model using original features.
But I'm in trouble of acquiring video features from "CUHK & ETH & SIAT Solution to ActivityNet Challenge 2016" model.
I want to get the original features used in your paper. (2048-dim I think, and whose length is not rescaled to 100)
Will you able to provide the 2048-dim original two stream features not rescaled?
Thank you.
Hi Pilhyeon
During feature extraction, we use "fc-action" layer, so the dimension of feature is 400 ( 200 + 200). I will upload feature zip file days later.
Hi Pilhyeon
I am uploading original feature in BaiduYun:
链接: https://pan.baidu.com/s/1871KzGL0KlVCgTBpMf_sag
提取码: h9st
The uploading procedure will finish in one or two hours. Please feel free to reopen the issue if you have furthur question.
Hello, @wzmsltw
First of all, Thanks to uploading the original feature!
However, downloading the file from Baidu Yun required the Baidu account.
Could you able to re-upload the same file to google drive for non-Badiu users?
Thank you
Hi @cominger , I am sorry that my google cloud disk does not has enough space for the feature. So you can sign up a baidu account or can you find any free net-disk?
Hi! Tianwei,@wzmsltw
I have the following questions I would like to ask you.
- When I extract the features of the 'fc-action' layer, I get the feature size (25, 10, 200), where 25 is the num_frame_per_video parameter in tsn. What does 10 mean? In the feature extraction process, did you set num_frame_per_video to the total number of frames in the video?
- In the readData function of the data_process.py file, how do I get the CSV format features in the spatial_dir and temporal_dir folders?
In the poolData function of the data_process.py file, what is the difference between the num_prop parameter and the num_frame_per_video parameter in tsn? - How do I get the 100*400 features for the TEM module?
Thank you very much! Looking forward to your reply!
@wzmsltw
About question 1, the link of eval_net.py here: https://github.com/yjxiong/temporal-segment-networks/blob/master/tools/eval_net.py
Should I save the 'frame_scores' of line92 to a separate CSV file (a video corresponds to a CSV file)?
About question 2,the link of data_process.py here: https://github.com/wzmsltw/BSN-boundary-sensitive-network.pytorch/blob/master/data/activitynet_feature_cuhk/data_process.py
Wish you a good time!
@wduo do you solve the problem?
@zys20172212265 have you solved the problem? thanks
@zys20172212265 Solved.
@wduo Since im just getting started in this filed , Could you show me more details about your question 3
@zys20172212265 Solved.
Hi! I'm confused about how to extract feature for every snippet? Does it mean that I need to use every snippet(16 frames as Tianwei do)as input of TSN? And the score of snippets is the output(feature)?