Multi-Stage Image Aesthetic Assessment via Chain-of-Thought Reasoning
Bibliographic record
Abstract
Image Aesthetic Assessment (IAA) is an crucial task in computer vision, aiming to quantify the aesthetic quality of images. Existing methods face two main challenges: neglecting the sequential modeling of human visual perception, and the fact that multi-attribute annotation is extremely time-consuming and labor-intensive. This letter proposes a novel Multi-stage IAA framework, leveraging Chain-of-Thought (CoT) reasoning and Multimodal Large Language Models (MLLMs). The framework designs a system that mimics the chain-like progression of human cognitive processing, with three dedicated modules: Low-level Stimulus Assessment (e.g., color harmony), Holistic Organizing Assessment (e.g., scene semantics), and High-level Perceiving Assessment (e.g., emotional resonance). These modules progressively analyze images, from basic visual features to high-level emotional understanding, by employing a step-by-step reasoning process characteristic of CoT. Furthermore, we explore an MLLM-oriented data transformation paradigm to convert multi-source IAA datasets into structured, Chain-of-Thought-compatible data, facilitating convenient attribute annotation. This enables the MLLMs to effectively learn and apply the CoT reasoning for aesthetic evaluation. Experiment on the PARA dataset achieves state-of-the-art results. Moreover, the framework exhibits generalization capabilities on the unseen FLICKR-AES dataset.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.007 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.003 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.002 | 0.003 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.004 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".