SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2410.00321 · NeurIPS · 2024

A Cat Is A Cat (Not A Dog!): Unraveling Information Mix-ups in Text-to-Image Encoders through Causal Analysis and Embedding Optimization

Chieh-Yun Chen, Chiang Tseng, Li-Wu Tsao, Hong-Han Shuai

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 7 functions out of this paper's own repositories and ran 5 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
basiclab/Unraveling-Information-Mix-ups canonical 4 of 6
basiclab/unraveling-information-mix-ups canonical 1 of 1
FunctionStatusWhere it lives
attn_map_tr2np Ran basiclab/Unraveling-Information-Mix-ups/utils/utils.py
code served (permissive licence) · get_code("4f1fe3d1f6b89580")
check_mixture Ran basiclab/Unraveling-Information-Mix-ups/eval_metrics/owlv2.py
code served (permissive licence) · get_code("99fc8c88c6520b92")
load_animals Ran basiclab/unraveling-information-mix-ups/data/gen_prompt.py
code served (permissive licence) · get_code("5feb9e1dddda5062")
resize_net_attn_map Ran basiclab/Unraveling-Information-Mix-ups/utils/utils.py
code served (permissive licence) · get_code("abb4ecbad6e0555f")
seed_everything Ran basiclab/Unraveling-Information-Mix-ups/utils/utils.py
code served (permissive licence) · get_code("dad22709a45ad4ab")
owlv2_eval Not yet run basiclab/Unraveling-Information-Mix-ups/eval_metrics/owlv2.py
code served (permissive licence) · get_code("346d093f84857a6b")
register_attention_control Not yet run basiclab/Unraveling-Information-Mix-ups/model/attentions_utils.py
code served (permissive licence) · get_code("ff0f7c7f0e3e3086")

Repositories linked to this paper

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

Abstract

This paper analyzes the impact of causal manner in the text encoder of text-to-image (T2I) diffusion models, which can lead to information bias and loss. Previous works have focused on addressing the issues through the denoising process. However, there is no research discussing how text embedding contributes to T2I models, especially when generating more than one object. In this paper, we share a comprehensive analysis of text embedding: i) how text embedding contributes to the generated images and ii) why information gets lost and biases towards the first-mentioned object. Accordingly, we propose a simple but effective text embedding balance optimization method, which is training-free, with an improvement of 125.42% on information balance in stable diffusion. Furthermore, we propose a new automatic evaluation metric that quantifies information loss more accurately than existing methods, achieving 81% concordance with human assessments. This metric effectively measures the presence and accuracy of objects, addressing the limitations of current distribution scores like CLIP's text-image similarities. The code is available: https://github.com/basiclab/Unraveling-Information-Mix-ups. 38th Conference on Neural Information Processing Systems (NeurIPS 2024).

For agents

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

get_harvested_code_for_paper("2410.00321")
get_code_for_paper("2410.00321")
have("2410.00321")

Connect an agent — have() is free.