two-stream-action-recognition
two-stream-action-recognition copied to clipboard
mohammed-elkomy
Dataset
Hello, do you have RGB images and optical flow images extracted from UCF101 and HMDB51? I have downloaded the images processed by Feichtenhofer using google colab but it seems the zip file is broken and I cannot get images. So could you share your data with me through google drive? Thanks a lot!
Unfortunately, I don't have them on my drive, but I believe there's something went wrong while downloading them, They are actually incomplete since Colab storage is now less the 70GB while the dataset for RGB images are about 33GB compressed so you need at least 120GB to decompress them, I suggest executing WGET command on your mounted drive space if you enough space on your drive this may work.
Step 1: Mount your drive Step 2: change directory to your drive Step 3: execute WGET in this directory if you have about 80GB free on your drive
Good Luck.
The problem comes when I extract them from zip file. Actually, I use WGET command on my colab and store the three ucf101_jpegs_256.zip.* file separately and then use CAT command to concat them to a single zip file, which is about 27G. However, "inflating: jpegs_256/v_Archery_g02_c07/frame000056.jpg error: zipfile read error" occurs when I unzip this 27G file.
Yes, you can't really decompress them on Colab, the instance doesn't have enough storage, you need at least 100 GB free storage (while decompressing a 1GB file you actually need at least 2GB, 1GB for the zipped and at least 1 GB for the decompressed form)
So you mean that I need more than 100 GB storage on colab while my 115 GB google drive doesn't work here?
You drive may work, but you need to
Step 1: Mount your drive Step 2: change working directory to your drive using import os os.chdir("/my drive/bla bla")
Step 3: execute WGET in this directory if you have about 80GB free on your drive !wget LINK_TO_FIRST_ZIP_FILE !wget LINK_TO_SECOND_ZIP_FILE !wget LINK_TO_THIRD_ZIP_FILE !CAT THEM >> all.zip !remove FIRST, SECOND, THIRD ZIP !unzip all.zip !rm all.zip
Yep, I have done the above steps but the error occurs when I carried out the "!unzip all.zip" (my command is "!unzip ucf101_jpegs_256.zip") in Step3.
Are you sure it's downloaded to your drive, and the Colab storage bar isn't full?
Hmm, I'm confused. I suggest deleting the zipped files and download, cat and decompress them again.
Yeah, I'm sure it's not full. I don't use GPU when preparing dataset and the avilable storage of Disk is about 77G
Yes. And I also tried "!jar xvf" command but it showed "Input/Output error".
Hmm, It seems really broken. My last idea is to wait a bit for Colab to sync with drive, It took me a few seconds to see them synced even for MB files, I suggest cleaning your drive and then download the 3 files then wait for a few minutes(I suggest 30 mins) then run the cat command and also wait again to get the compressed file and then decompress and have a lunch break :D
I say that since the links are mostly not broken.
I really want to help you :(
I didn't encounter those issues when working on it since Colab py3-GPU instance had about 300GBs !
How to get 300 GBs? Is it Colab Pro? Here, I encounter a new problem: "Buffered data was truncated after reaching the output size limit."
How to get 300 GBs? Is it Colab Pro? That was the norm before :(
Here, I encounter a new problem: "Buffered data was truncated after reaching the output size limit." Drive API has limits on Quota, I think you exceeded the limits :(
I think it's a little bit difficult for me to finish my task with colab and I may be supposed to use a Linux machine.