[data request] Downsampled ImageNet #45
Comments
|
As this is a good first issue would like to take this up 😀 |
|
There are 2 version of this dataset 32 * 32 images and 64 * 64 images should both be done ? |
|
That's great @tabshaikh, thank you! Yes, I think we should have both using "heavy" configuration, but you can start with just one. |
|
I'll assign the issue to you as soon as you accept the collaborator invite! 😃 |
|
@rsepassi invite accepted would have a pr soon :) |
|
Sounds good! Thank you! |
|
@rsepassi I would love to collaborate with @tabshaikh and make a pull request for Downsampled ImageNet. Please assign it to me also. |
|
Hi @anupam-tripathi, thanks for your interest! Are you already working directly with @tabshaikh? If not, let's give him a chance to get a PR in. If he'd like the help, then please do work together to get something in! |
|
No, I have not joined him till now but will surely contact him personally. |
|
@anupam-tripathi I would love to collaborate with you but the pr is almost done with a few changes left to do and hopefully, I have added the dataset correctly @rsepassi I would do a pr till mid next week as I will be going for an ML hackathon during the weekend, I had some question too would ask in the draft pr |
|
Sounds good. Thanks!
…On Fri, Mar 1, 2019 at 3:08 PM Tabish Shaikh ***@***.***> wrote:
@anupam-tripathi <https://github.com/Anupam-tripathi> I would love to
collaborate with you but the pr is almost done with a few changes left to
do and hopefully, I have added the dataset correctly @rsepassi
<https://github.com/rsepassi> I would do a pr till mid next week as I
will be going for an ML hackathon during the weekend, I had some question
too would ask in the draft pr
Cheers :)
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#45 (comment)>,
or mute the thread
<https://github.com/notifications/unsubscribe-auth/ABEGW1o0D_TGuZv0C1PAarkIl1Q9BwFcks5vSbLtgaJpZM4au74x>
.
|
|
Ya, surely I will prove to be a good teammate. |
|
@rodrigob @rsepassi the link for http://image-net.org/small/download.php does not contain the whole dataset of the downsampled imagenet nor does it contain the labels. |
|
@tabshaikh - Yes, this dataset has only a subset of subsampled imagenet images and does NOT have labels. This is on purpose - as it was used for autoregressive algorithms, that were generating the output images (rather than trying to predict the class). Please download from the official link rather than from side-ones. |
|
@cyfra okay cool |
|
The idea of this ticket was to create a smaller version of imagenet that is small enough so that most people can prototype and experiment without having to worry about download time or disk-space. I would suggest to go for 32x32 and 64x64 versions; ideally with labels so that supervised training (à la MNIST and CIFAR) can also be used. |
|
@rodrigob okay thanks :) |
|
Any thoughts on doing the 128x128 version while you're at it? |
|
Joel, looks like there are only 2 versions listed, 32 and 64: http://image-net.org/small/download.php |
|
@joel-shor Do you have a link to 128x128? |
|
@rsepassi no there are 4 versions actually 8x8, 16x16, 32x32, 64x64 here http://image-net.org/download-images. The link which you pointed out is incomplete as there is no labels for the same. |
|
Sorry for the delay. The 128x128 imagenet, which is used in a number of state-of-the-art GANs (such as Self Attention GAN), can be found here: https://github.com/openai/improved-gan/blob/master/imagenet/convert_imagenet_to_records.py If you were able to turn this in to a TFDS data set, you would be a hero! |
|
@tabshaikh Have you moved on from this? |
|
Has been added with #613. Closing this now |
Folks who would also like to see this dataset in
tensorflow/datasets, please +1/thumbs-up so the developers can know which requests to prioritize.The text was updated successfully, but these errors were encountered: