Datasets:

ArXiv:
License:
tianbaoxiexxx commited on
Commit
241e85c
·
verified ·
1 Parent(s): 90ed6fe

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +3 -1
README.md CHANGED
@@ -15,7 +15,9 @@ The JEDI Dataset consists of four carefully designed categories:
15
 
16
  This repository includes all the textures and images for these components.
17
 
18
- Additionally, JEDI processes and improves the data from [AGUVIS](https://github.com/xlang-ai/aguvis) and [Desktop domain from OS-ATLAS](https://huggingface.co/datasets/OS-Copilot/OS-Atlas-data/tree/main/desktop_domain), calling it `AGUVIS++`. This repository contains the texture portion of `AGUVIS++`. For images, please refer to the [aguvis original repository](https://huggingface.co/collections/ranpox/aguvis-unified-pure-vision-gui-agents-6764e2bc343c62af95c209d8) and [os-atlas original repository](https://huggingface.co/datasets/OS-Copilot/OS-Atlas-data/tree/main/desktop_domain), please download them and align with the image path for usage.
 
 
19
 
20
  Project Page: https://osworld-grounding.github.io
21
  GitHub Repository: https://github.com/xlang-ai/OSWorld-G
 
15
 
16
  This repository includes all the textures and images for these components.
17
 
18
+ Additionally, JEDI processes and improves the data from [AGUVIS](https://github.com/xlang-ai/aguvis) and [Desktop domain from OS-ATLAS](https://huggingface.co/datasets/agnet/osatlas), calling it `AGUVIS++`. This repository contains the texture portion of `AGUVIS++`. For images, please refer to the [aguvis original repository](https://huggingface.co/collections/ranpox/aguvis-unified-pure-vision-gui-agents-6764e2bc343c62af95c209d8) and [os-atlas original repository](https://huggingface.co/datasets/OS-Copilot/OS-Atlas-data/tree/main/desktop_domain), please download them and align with the image path for usage.
19
+
20
+ **NOTE**: Before you use this dataset, make sure you understand the logic of absolute coordinates for Qwen2.5-VL. This dataset is set with the image processor max tokens to be 2700, a.k.a max_pixels=2700x14x14x2x2 , make sure you also follow it in your training procedure, otherwise the performance will not be as expected.
21
 
22
  Project Page: https://osworld-grounding.github.io
23
  GitHub Repository: https://github.com/xlang-ai/OSWorld-G