Fake Paper, the layout is not reproducible at all

xbmxb commented 4 days ago

The layouts are provided in the original versions of AitW and META-GUI.

We have provided preprocessing code and preprocessed META-GUI data. The processed AitW data follows the same format as the preprocessed META-GUI data. We are currently working on uploading the processed AitW training data, which is quite large in size.

sugarandgugu commented 1 day ago

怎么被质疑造假了哈哈

xbmxb commented 1 day ago

Layout在 AitW 和 META-GUI 的原始版本中提供，前者提供Iconnet预测的bbox作为layout，后者提供xml文件。具体请参见二者原始仓库，https://github.com/google-research/google-research/tree/master/android_in_the_wild，https://x-lance.github.io/META-GUI-Leaderboard/。

我们基于原数据集标注进行处理，取中心点作为位置标注。此路径下有数据处理的代码，以及META-GUI的处理后版本（AitW数据较大，会尽快传好），请做参考。https://github.com/xbmxb/CoCo-Agent/tree/main/data

===================================================================

Layout is provided in the original versions of AitW and META-GUI. The former provides the bbox predicted by Iconnet as layout, and the latter provides an xml file. For details, please refer to the original repositories of the two, https://github.com/google-research/google-research/tree/master/android_in_the_wild, https://x-lance.github.io/META-GUI-Leaderboard/.

We process based on the original dataset annotations and take the center as the position annotation. There is data processing code and the processed version of META-GUI in this path (AitW data is large and will be uploaded as soon as possible), please refer to it. https://github.com/xbmxb/CoCo-Agent/tree/main/data

Ablus-Bang commented 1 day ago

Thanks for the reply, we'll keep waiting

xbmxb commented 9 hours ago

You can directly use this file (https://github.com/xbmxb/CoCo-Agent/blob/main/data/fetch_dataset_for_t5_blipv2_google_pre10.py) to generate our training and test data based on the original AitW and META-GUI. And you can also use our META-GUI data first.

Ablus-Bang commented 9 hours ago

First, I simply execute the program python fetch_dataset_for_t5_blipv2_google_pre10.py I was told that I didn't have the module android_in_the_wild, which made me think that the author probably didn't really want to open source it. There's no hint where to install this environment, but I found this module in google_research anyway I then proceeded to execute it, alerting me to KeyError: 'validation', so I abandoned the command

Then I found a commented instruction in the code that： python fetch_dataset_for_t5_blipv2_google_pre10.py --split_file "dataset/splits/standard.json" --output_dir "dataset/owl/google_apps_parsed_episode_owl_pre10" --dataset google_apps

Again this has no meaningful results, I'm wondering if the authors are dealing with censorship by not providing either the environment or the instructions to execute it, open source but only half open source

On Wed, Nov 6, 2024 at 1:04 AM Xinbei @.***> wrote:

You can directly use this file ( https://github.com/xbmxb/CoCo-Agent/blob/main/data/fetch_dataset_for_t5_blipv2_google_pre10.py) to generate our training and test data based on the original AitW and META-GUI. And you can also use our META-GUI data first.

— Reply to this email directly, view it on GitHub https://github.com/xbmxb/CoCo-Agent/issues/5#issuecomment-2457265517, or unsubscribe https://github.com/notifications/unsubscribe-auth/BLSU7JCCDZRS6XA6ESSEHTLZ7DF6LAVCNFSM6AAAAABQ6VUVTKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDINJXGI3DKNJRG4 . You are receiving this because you authored the thread.Message ID: @.***>

xbmxb / CoCo-Agent

Fake Paper, the layout is not reproducible at all #5