SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2409.14818 · 2024

MobileVLM: A Vision-Language Model for Better Intra- and Inter-UI Understanding

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 18 functions out of this paper's own repositories and ran 13 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
xiaomi/mobilevlm canonical 13 of 18
FunctionStatusWhere it lives
calculate_iou Ran xiaomi/mobilevlm/test_515/count_aver_iou.py
pointer only (licence: NOASSERTION) · get_code("3aac5dfd1627a360")
dict2dict Ran xiaomi/mobilevlm/corpus/list_prefix.py
pointer only (licence: NOASSERTION) · get_code("e4b5d64062a91b7f")
dict2txt Ran xiaomi/mobilevlm/corpus/list_prefix.py
pointer only (licence: NOASSERTION) · get_code("48c980f8b78480c1")
dictcount Ran xiaomi/mobilevlm/corpus/list_prefix.py
pointer only (licence: NOASSERTION) · get_code("b2e1feaa54352576")
gpt_usage Ran xiaomi/mobilevlm/corpus/chatgpt.py
pointer only (licence: NOASSERTION) · get_code("e479638f07e3d5f2")
load_dict Ran xiaomi/mobilevlm/corpus/bm25.py
pointer only (licence: NOASSERTION) · get_code("e0f0dbdcd40e0123")
load_json Ran xiaomi/mobilevlm/corpus/bm25.py
pointer only (licence: NOASSERTION) · get_code("1c84aeb1de479bcc")
load_txt Ran xiaomi/mobilevlm/corpus/bm25.py
pointer only (licence: NOASSERTION) · get_code("1aa5d2a1084ca8a3")
local_image_to_data_url Ran xiaomi/mobilevlm/test_515/gpt.py
pointer only (licence: NOASSERTION) · get_code("829b3a29017bf6be")
preprocess Ran xiaomi/mobilevlm/finetune.py
pointer only (licence: NONE) · get_code("8b733c9fc590e4c1")
process_html Ran xiaomi/mobilevlm/test_515/build_test_inout_put.py
pointer only (licence: NOASSERTION) · get_code("d4b2c3b54965a829")
read_json_file Ran xiaomi/mobilevlm/corpus/xml_test_22.py
pointer only (licence: NOASSERTION) · get_code("707a0ecb9f3c6f2d")
scale_coordinates Ran xiaomi/mobilevlm/test_515/count_aver_iou.py
pointer only (licence: NOASSERTION) · get_code("a8686802123b33d2")
any_tree_to_html Not yet run xiaomi/mobilevlm/xml_to_html.py
pointer only (licence: NOASSERTION) · get_code("2f9f1293699146bc")
chatgpt Not yet run xiaomi/mobilevlm/corpus/chatgpt.py
pointer only (licence: NOASSERTION) · get_code("ff85b6fede884ce7")
extract_coordinates Not yet run xiaomi/mobilevlm/test_515/count_aver_iou.py
pointer only (licence: NOASSERTION) · get_code("7261ce7a4e26085e")
gpt_4v_actionspace Not yet run xiaomi/mobilevlm/test_515/gpt.py
pointer only (licence: NOASSERTION) · get_code("b6659c22b1479358")
gpt_4v_ref Not yet run xiaomi/mobilevlm/test_515/gpt.py
pointer only (licence: NOASSERTION) · get_code("e513b28efe0be636")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Recently, mobile AI agents based on VLMs have been gaining increasing attention. These works typically utilize VLM as a foundation, fine-tuning it with instruction-based mobile datasets. However, these VLMs are typically pre-trained on general-domain data, which often results in a lack of fundamental capabilities specific to the mobile domain. Therefore, they may struggle to recognize specific UI elements and understand intra-UI fine-grained information. In addition, the current fine-tuning task focuses on interacting with the most relevant element for the given instruction. These fine-tuned VLMs may still ignore the relationships between UI pages, neglect the roles of elements in page transitions and lack inter-UI understanding. To address issues, we propose a VLM called MobileVLM, which includes two additional pre-training stages to enhance both intra- and inter-UI understanding. We defined four UI-based pre-training tasks, enabling the model to better perceive fine-grained elements and capture page transition actions. To address the lack of mobile pre-training data, we built a large Chinese mobile dataset Mobile3M from scratch, which contains 3 million UI pages, and real-world transition actions, forming a directed graph structure. Experimental results show MobileVLM excels on both our test set and public mobile benchmarks, outperforming existing VLMs.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2409.14818")
get_code_for_paper("2409.14818")
have("2409.14818")

Connect an agent — have() is free.