Summary of Revisiting Multi-modal Llm Evaluation, by Jian Lu et al.
Revisiting Multi-Modal LLM Evaluation
by Jian Lu, Shikhar Srivastava, Junyu Chen, Robik Shrestha, Manoj Acharya, Kushal Kafle, Christopher Kanan
First submitted to arxiv on: 9 Aug 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper pioneers evaluating recent multi-modal large language models (MLLMs) on datasets designed to address weaknesses in earlier ones. The authors assess the performance of six MLLMs (LLaVA 1.5, LLaVA-NeXT, BLIP2, InstructBLIP, GPT-4V, and GPT-4o) on three visual question answering (VQA) datasets: TDIUC, which permits fine-grained analysis; TallyQA, with simple and complex counting questions; and DVQA, requiring optical character recognition for chart understanding. Additionally, the authors study VQDv1, a dataset that requires identifying all image regions satisfying a given query. The experiments reveal weaknesses in many MLLMs not previously reported. The code is integrated into the LAVIS framework, enabling rapid assessment of future MLLMs. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper evaluates how well big AI models can answer questions about pictures and videos. Right now, these models are very good at some things, but bad at others. The authors created new tests to see where these models go wrong and what they need to do better. They looked at six different models on four different tasks, like counting objects in a picture or understanding charts. They found that many of the models struggled with certain types of questions, which is important to know so we can make them better. |
Keywords
» Artificial intelligence » Gpt » Multi modal » Question answering
