csailvision / ade20k Goto Github PK

Using ObjectClassMasks = (R/10).astype(np.int32)*256+(G.astype(np.int32)); I got the id 165 for a particular pixel by which index_ade20k['objectnames'][165] gives 'bed column'. However according to the accompanying json to this image, the object name for 165 should be 'bed'. It seems that index_ade20k['objectnames'] is one behind. index_ade20k[164] gives the correct name 'bed'.

I request for the dataset to train a model to use segmentation as a signal for some fusion, I'm from Rutgers University, I have requested for access on the portal, not sure how long that takes, requesting here for confirmation of need of the dataset'

Could you please point me to the mapping of RGB values to object labels? Or is there no such mapping in one place. Am I to somehow make use the individual instance masks to get the object labels per pixel?

Dear Authors,

Thank for the great work to create and maintain the dataset.
I would like to have access to the download link but the approval is still pending.
I checked around but there seemed to be no information about how long the waiting period would be.
Therefore I post here to ask for help.
My account is angel33468.

Thanks in advance!

Hello! I try to get an access for your dataset from site, but I still see such message when log in:

Your access to the dataset is pending approval. When you are approved, a link will appear in this page

Can you please accept my request? My login is DaraOrange, I'm a student of Computer Science from NRU HSE and I need this dataset for my research in CV very much.

Hi
When I try to Load a segmentation object as provided by the ade20k_starter.ipynb I receive the following error. Is my dataset incomplete or am I missing something?

FileNotFoundError: [Errno 2] No such file or directory: 'D:\deft and tact\vickyskarthik88_6d602e54\ADE20K_2021_17_01/ADE20K_2021_17_01/images/ADE/training/urban/street/ADE_train_00016869_seg.png'

Thanks

I registered with academic email (login shkarupaalex) about a month ago and still waiting for approval.
Could you please say how long this action takes?

Thanks for curating this dataset. It's fun to work with.

I have a question regarding the intersection between the validation set of ADE20k and the train set of Places365 (standard) with roughly 1.8M images. Do you know by any chance whether any images of the train set of Places365(standard) have ended up in the the validation set of ADE20k? I looked through your papers, but I could only find the following sentence

Images come from the LabelMe [30], SUN datasets [34], and Places [38] and were selected to cover the 900 scene categories defined in the SUN database. Images were annotated by a single expert worker using the LabelMe interface [30].

in Semantic Understanding of Scenes through the ADE20K Dataset, which leaves this question open. It would be great if you could clarify this.

Thanks for sharing this wonderful work!

I am planning to use your dataset to indoor object detection and use them for location rendering, could I have dataset access? My account is yhdl001 and I already submitted the request through your website.

Thanks for sharing this wonderful work! My account is macaroniz and already submitted the request through your website, could you please take a look and share the download link with me?

Thanks a lot!

Hi thanks for the wonderful dataset,
I am wondering, is there any way to determine whether a category is a "thing" (countable) or a "stuff" (uncountable) other than go through objectPresence? Even if I did this, some countable objects are still categorized as "uncountable" because they just appear once in the whole dataset (like abacus, airport cart,...)

Thanks
Dang

As the title says, where can I download ade20k_instance_imgCatIds.json? I can't find it below.

MIT Scene Parsing Benchmark http://sceneparsing.csail.mit.edu/

Hi,is it possible to filter the ADE20K dataset in such a way that only images that have a specific class are included
For example in the dataset there are 318 images corresponding to the class bridge.I would like to retain only these images and ignore the rest of them.Is that possible,and if yes how would you suggest i go about it.

I am currently working with semantic segmentation on indoor and outdoor scene and would like to use the ADE20K dataset for training. Please, accept my request to access the dataset!

Thanks for sharing this wonderful work!

I am planning to use your dataset to do some research, could I have dataset access? My account is wzq016 and already submitted the request through your website.

Hi,
I will start in few days my Computer Vision project at university. The purpose is to segment indoor rooms. I would like to use the ADE20K dataset for training my model. my login is: igghi2, I did the request few weeks ago.
Is it possible to accept my request to the dataset ?
Thanks.

Hi @xavierpuigf,
thank you very much for your wonderful work! I would like to use your dataset to train my model used in a university project. My account name is mdnguyen. I'd appreciate it if you could take some time to approve my request to access the dataset. Thank you very much!

Thanks,
Dang Nguyen

Hi, I registered on your website with the user name RBS1024, and the email is an edu one. I am a student interested in segmentation, could you please give me the approval? Thanks!

Thanks for sharing this wonderful work! My account is jimengying and already submitted the request through your website, could you please take a look and share the download link with me?

Thanks a lot!

Thanks for sharing this wonderful work!

I am planning to use your datasets to do some research on visual affordance, could I have datasets access?
My account is wangyuchen and already submitted the request through your website.

when I click Register full dataset, the website navigates an empty page!

I am currently working on indoor scene recognition and would like to use the ADE20K dataset for training my model
Kindly accept my request to access the dataset

Hi:

Could you provide the groundtruth boundary for ADE20K dataset?

Thanks a lot.

Is SceneParsing150 the original snapshot of ADE20k?

Multiple papers refer to ADE20k has having 20210 training images, which matches this description of SceneParsing150. Can someone clarify if this is the case?

This is important if we want to compare our results with this leaderboard. Confirmation will be greatly appreciated.

Thank you.

Hi, I would like to know if there are some way to get old version of ADE20k, which I believe is 2017 version.

Upon this issue from mmsegmentation, it states it use old version but currently could not found the link for the download.

It would be really nice if you could provide script to convert to that old version.

Thanks.

Hi, I have submitted my application of data access on the website, my account is "hktech".
I use the dataset for research only.

Hi! The ADE20K dataset is very helpful to my work! But I wonder where can I get the full stuff class (non discrete object and object part) name list.

I have read the paper (listed in https://groups.csail.mit.edu/vision/datasets/ADE20K/), and have searched for days. I only found a Google Document (https://docs.google.com/spreadsheets/d/1se8YEtb2detS7OuPE86fXGyD269pMycAWe2mtKUj2W8/edit#gid=0, mentioned in lllyasviel/ControlNet#172), but it seems like a list which is only part of this dataset (SceneParse150?).

I'll appreciate if you could help me, thanks!

Need access for bench-marking. All the help is appreciated

Hi,

I have one question. As is known, the ADE20K dataset is publicly available semantic segmentation dataset. As I have seen some papers conduct semantic segmentation using boundary constraint or prediction, I wonder to know whether the ADE20K dataset is also labeled with groundtruth boundary.

Thanks.

Hi, thanks for your work!
I want to use the dataset for a university project but it seems that the approval on the website wont come in. My account name is GabrieleFerrario and my display name is Gabry.

Thanks,
Gabriele

I noticed that the way to get the class mask is:
ObjectClassMasks = (R/10).astype(np.int32)*256+(G.astype(np.int32))

Is there a similar way to get the instance information or instance mask from the RGB channels for both object level and part level segmentations?

Thanks!

I am trying to transfer ADE20k dataset to roboflow account. What format is ADE20k dataset json format. Can we convert it to COCO json format?

Hi, I want to access ADE20K dataset on https://groups.csail.mit.edu/vision/datasets/ADE20K.

Could you check my request and approve? Thank you.

Thanks for sharing this wonderful work! My account is haowang. I submitted a request through the website several weeks ago and I still don't receive any feedback. Would you mind taking some time to accept my request?

Thanks in advance!

We're working on a segmentation problem for the parts in the object we request you to provide the access for the mail [email protected] ( IIIT Bangalore College Name)
Can you approve the dataset fast.
Thank you so much

Hi! I put in a request for access to the dataset about a week ago. I'm still waiting with my request pending approval. Can you please review my request and hopefully approve?

Thanks!

Hello, I want to use the dataset for a university project but it seems that the approval on the website wont come in. My account name is RobertCaliman15.

Hello, I requested access to this dataset with the username massel1 under an email ending with illinois.edu. I plan on using this dataset for a research project, would it be possible to accept my request?

Hi, I am a student interested in segmentation and registered on the website, yet fail to log into the website. Could you take a look?
My user name is feinanshan, and my email ends up with bu.edu. Thanks!

Thanks for sharing this wonderful work! My account is ssaily and already submitted the request through your website, could you please take a look and share the download link with me?

Thanks a lot!

I will start next week my research for my master degree. The purpose is to detect the parts on cars. I would like to use the ADE20K dataset for training my model. my login is: philbl, I did the request few weeks ago.
Is it possible to accept my request to the dataset ?
Thanks,