ConceptMix: A Compositional Image Generation Benchmark with Controllable Difficulty
- URL: http://arxiv.org/abs/2408.14339v1
- Date: Mon, 26 Aug 2024 15:08:12 GMT
- Title: ConceptMix: A Compositional Image Generation Benchmark with Controllable Difficulty
- Authors: Xindi Wu, Dingli Yu, Yangsibo Huang, Olga Russakovsky, Sanjeev Arora,
- Abstract summary: ConceptMix is a scalable, controllable, and customizable benchmark.
It automatically evaluates compositional generation ability of Text-to-Image (T2I) models.
It reveals that the performance of several models, especially open models, drops dramatically with increased k.
- Score: 52.15933752463479
- License: http://creativecommons.org/licenses/by/4.0/
- Abstract: Compositionality is a critical capability in Text-to-Image (T2I) models, as it reflects their ability to understand and combine multiple concepts from text descriptions. Existing evaluations of compositional capability rely heavily on human-designed text prompts or fixed templates, limiting their diversity and complexity, and yielding low discriminative power. We propose ConceptMix, a scalable, controllable, and customizable benchmark which automatically evaluates compositional generation ability of T2I models. This is done in two stages. First, ConceptMix generates the text prompts: concretely, using categories of visual concepts (e.g., objects, colors, shapes, spatial relationships), it randomly samples an object and k-tuples of visual concepts, then uses GPT4-o to generate text prompts for image generation based on these sampled concepts. Second, ConceptMix evaluates the images generated in response to these prompts: concretely, it checks how many of the k concepts actually appeared in the image by generating one question per visual concept and using a strong VLM to answer them. Through administering ConceptMix to a diverse set of T2I models (proprietary as well as open ones) using increasing values of k, we show that our ConceptMix has higher discrimination power than earlier benchmarks. Specifically, ConceptMix reveals that the performance of several models, especially open models, drops dramatically with increased k. Importantly, it also provides insight into the lack of prompt diversity in widely-used training datasets. Additionally, we conduct extensive human studies to validate the design of ConceptMix and compare our automatic grading with human judgement. We hope it will guide future T2I model development.
Related papers
- Is What You Ask For What You Get? Investigating Concept Associations in Text-to-Image Models [24.851041038347784]
Text-to-image (T2I) models are increasingly used in real-life applications.
There is a growing need to audit these models to ensure that they generate desirable, task-appropriate images.
We propose Concept2Concept, a framework where we characterize conditional distributions of vision language models.
arXiv Detail & Related papers (2024-10-06T21:42:53Z) - Non-confusing Generation of Customized Concepts in Diffusion Models [135.4385383284657]
We tackle the common challenge of inter-concept visual confusion in compositional concept generation using text-guided diffusion models (TGDMs)
Existing customized generation methods only focus on fine-tuning the second stage while overlooking the first one.
We propose a simple yet effective solution called CLIF: contrastive image-language fine-tuning.
arXiv Detail & Related papers (2024-05-11T05:01:53Z) - MC$^2$: Multi-concept Guidance for Customized Multi-concept Generation [59.00909718832648]
We propose MC$2$, a novel approach for multi-concept customization.
By adaptively refining attention weights between visual and textual tokens, our method ensures that image regions accurately correspond to their associated concepts.
Experiments demonstrate that MC$2$ outperforms training-based methods in terms of prompt-reference alignment.
arXiv Detail & Related papers (2024-04-08T07:59:04Z) - Gen4Gen: Generative Data Pipeline for Generative Multi-Concept
Composition [47.07564907486087]
Recent text-to-image diffusion models are able to learn and synthesize images containing novel, personalized concepts.
This paper tackles two interconnected issues within this realm of personalizing text-to-image diffusion models.
arXiv Detail & Related papers (2024-02-23T18:55:09Z) - Visual Concept-driven Image Generation with Text-to-Image Diffusion Model [65.96212844602866]
Text-to-image (TTI) models have demonstrated impressive results in generating high-resolution images of complex scenes.
Recent approaches have extended these methods with personalization techniques that allow them to integrate user-illustrated concepts.
However, the ability to generate images with multiple interacting concepts, such as human subjects, as well as concepts that may be entangled in one, or across multiple, image illustrations remains illusive.
We propose a concept-driven TTI personalization framework that addresses these core challenges.
arXiv Detail & Related papers (2024-02-18T07:28:37Z) - Textual Localization: Decomposing Multi-concept Images for
Subject-Driven Text-to-Image Generation [5.107886283951882]
We introduce a localized text-to-image model to handle multi-concept input images.
Our method incorporates a novel cross-attention guidance to decompose multiple concepts.
Notably, our method generates cross-attention maps consistent with the target concept in the generated images.
arXiv Detail & Related papers (2024-02-15T14:19:42Z) - ConceptBed: Evaluating Concept Learning Abilities of Text-to-Image
Diffusion Models [79.10890337599166]
We introduce ConceptBed, a large-scale dataset that consists of 284 unique visual concepts and 33K composite text prompts.
We evaluate visual concepts that are either objects, attributes, or styles, and also evaluate four dimensions of compositionality: counting, attributes, relations, and actions.
Our results point to a trade-off between learning the concepts and preserving the compositionality which existing approaches struggle to overcome.
arXiv Detail & Related papers (2023-06-07T18:00:38Z) - Break-A-Scene: Extracting Multiple Concepts from a Single Image [80.47666266017207]
We introduce the task of textual scene decomposition.
We propose augmenting the input image with masks that indicate the presence of target concepts.
We then present a novel two-phase customization process.
arXiv Detail & Related papers (2023-05-25T17:59:04Z)
This list is automatically generated from the titles and abstracts of the papers in this site.
This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.