OPERA
by shikiw
[CVPR 2024 Highlight] OPERA: Alleviating Hallucination in Multi-Modal Large Language Models via Over-Trust Penalty and Retrospection-Allocation
AI summary
Token penalty
A method to alleviate hallucination in large language models by penalizing over-trust and re-allocation of tokens during decoding
- stars
- 293
- forks
- 27
- watching
- 3
Similar projects
Found by comparing what the projects do, not just their names.
Benchmark
An image-context reasoning benchmark designed to challenge large vision-language models and help improve their accuracy
Account hijack test
Proof-of-concept project demonstrating a GitHub account hijacking vulnerability using Baidu and a specific security testing methodology.
Hallucination tester
Evaluates and mitigates hallucinations in multimodal large language models
Hallucination mitigation
A research project focused on mitigating hallucinations in large multi-modal models by improving instruction tuning through robust training methods.
Hallucination corrector
A method to correct hallucinations in multimodal large language models without requiring retraining
Model validator
Analyzing and mitigating object hallucination in large vision-language models to improve their accuracy and reliability.
Hallucination mitigation
Improving multimodal hallucination mitigation in EOS decision-making by selectively supervising training data
Object detector
An evaluation framework for detecting object hallucinations in vision-language models
Perception adapter
An adapter for improving large language models at object-level perception tasks with auxiliary perception modalities
Region-of-Interest Training
Training and deploying large language models on computer vision tasks using region-of-interest inputs
Token representation refinement
Improves pre-trained language models by encouraging an isotropic and discriminative distribution of token representations.
Model debiasing
Debiasing techniques to minimize hallucinations in large visual language models
Hallucination detector
An evaluation framework for assessing the performance of large language models on question-answering tasks with hallucination detection
Token stripper
A tool that weakens antivirus and EDR products by deleting process token privileges and lowering integrity level to untrusted.
Hallucination fixer
An implementation of a method to mitigate hallucinations in large language models using visual re-tracing
