Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Result] Update Evaluation Results #60

Merged
merged 8 commits into from
Jan 22, 2024
Merged

[Result] Update Evaluation Results #60

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
6c858d4
update MME, SEEDBench
kennymckormick Jan 20, 2024
8ee2eee
update results
kennymckormick Jan 20, 2024
8ac6e8d
update LLaVABench
kennymckormick Jan 21, 2024
941a519
fix
kennymckormick Jan 21, 2024
412c67c
update AI2D accuracy
kennymckormick Jan 21, 2024
8e046e6
update LLaVABench
kennymckormick Jan 21, 2024
641aaac
update README
kennymckormick Jan 21, 2024
2a079b3
update teaser link
kennymckormick Jan 22, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 5 additions & 6 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
![LOGO](https://github-production-user-asset-6210df.s3.amazonaws.com/34324155/295443340-a300f073-4995-48a5-af94-495141606cf7.jpg)
![LOGO](http://opencompass.openxlab.space/utils/MMLB.jpg)
<div align="center"><b>A Toolkit for Evaluating Large Vision-Language Models. </b></div>
<div align="center"><br>
<a href="https://opencompass.org.cn/leaderboard-multimodal">🏆 Learderboard </a> •
Expand All @@ -9,16 +9,15 @@
<a href="#%EF%B8%8F-citation">🖊️Citation </a>
<br><br>
</div>

**VLMEvalKit** (the python package name is **vlmeval**) is an **open-source evaluation toolkit** of **large vision-language models (LVLMs)**. It enables **one-command evaluation** of LVLMs on various benchmarks, without the heavy workload of data preparation under multiple repositories. In VLMEvalKit, we adopt **generation-based evaluation** for all LVLMs (obtain the answer via `generate` / `chat` interface), and provide the evaluation results obtained with both **exact matching** and **LLM(ChatGPT)-based answer extraction**.

## 🆕 News

- **[2024-01-14]** We have supported [**LLaVABench (in-the-wild)**](https://huggingface.co/datasets/liuhaotian/llava-bench-in-the-wild).
- **[2024-01-21]** We have updated results for [**LLaVABench (in-the-wild)**](/results/LLaVABench.md) and [**AI2D**](/results/AI2D.md).
- **[2024-01-14]** We have supported [**AI2D**](https://allenai.org/data/diagrams) and provided the [**script**](/scripts/AI2D_preproc.ipynb) for data pre-processing. 🔥🔥🔥
- **[2024-01-13]** We have supported [**EMU2 / EMU2-Chat**](https://github.com/baaivision/Emu) and [**DocVQA**](https://www.docvqa.org). 🔥🔥🔥
- **[2024-01-11]** We have supported [**Monkey**](https://github.com/Yuliang-Liu/Monkey). 🔥🔥🔥
- **[2024-01-09]** The performance numbers on our official multi-modal leaderboards can be downloaded in json files: [**MMBench Leaderboard**](http://opencompass.openxlab.space/utils/MMBench.json), [**OpenCompass Multi-Modal Leaderboard**](http://opencompass.openxlab.space/utils/MMLB.json). We also add a [notebook](scripts/visualize.ipynb) to visualize these results.🔥🔥🔥
- **[2024-01-09]** The performance numbers on our official multi-modal leaderboards can be downloaded in json files: [**MMBench Leaderboard**](http://opencompass.openxlab.space/utils/MMBench.json), [**OpenCompass Multi-Modal Leaderboard**](http://opencompass.openxlab.space/utils/MMLB.json). We also added a [**notebook**](scripts/visualize.ipynb) to visualize these results.🔥🔥🔥
- **[2024-01-03]** We support **ScienceQA (Img)** (Dataset Name: ScienceQA_[VAL/TEST], [**eval results**](results/ScienceQA.md)), **HallusionBench** (Dataset Name: HallusionBench, [**eval results**](/results/HallusionBench.md)), and **MathVista** (Dataset Name: MathVista_MINI, [**eval results**](/results/MathVista.md)). 🔥🔥🔥
- **[2023-12-31]** We release the [**preliminary results**](/results/VQA.md) of three VQA datasets (**OCRVQA**, **TextVQA**, **ChatVQA**). The results are obtained by exact matching and may not faithfully reflect the real performance of VLMs on the corresponding task.

Expand Down Expand Up @@ -46,9 +45,9 @@
| [**OCRVQA**](https://ocr-vqa.github.io) | OCRVQA_[TESTCORE/TEST] | ✅ | ✅ | [**VQA**](/results/VQA.md) |
| [**TextVQA**](https://textvqa.org) | TextVQA_VAL | ✅ | ✅ | [**VQA**](/results/VQA.md) |
| [**ChartQA**](https://github.com/vis-nlp/ChartQA) | ChartQA_VALTEST_HUMAN | ✅ | ✅ | [**VQA**](/results/VQA.md) |
| [**AI2D**](https://allenai.org/data/diagrams) | AI2D | ✅ | ✅ | [**AI2D**](/results/AI2D.md) |
| [**LLaVABench**](https://huggingface.co/datasets/liuhaotian/llava-bench-in-the-wild) | LLaVABench | ✅ | ✅ | [**LLaVABench**](/results/LLaVABench.md) |
| [**DocVQA**](https://www.docvqa.org) | DocVQA_VAL | ✅ | ✅ | |
| [**AI2D**](https://allenai.org/data/diagrams) | AI2D | ✅ | ✅ | |
| [**LLaVABench**](https://huggingface.co/datasets/liuhaotian/llava-bench-in-the-wild) | LLaVABench | ✅ | ✅ | |
| [**Core-MM**](https://github.com/core-mm/core-mm) | CORE_MM | ✅ | | |

**Supported API Models**
Expand Down
39 changes: 39 additions & 0 deletions results/AI2D.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,39 @@
# AI2D Evaluation Results

> During evaluation, we use `GPT-3.5-Turbo-0613` as the choice extractor for all VLMs if the choice can not be extracted via heuristic matching. **Zero-shot** inference is adopted.

## AI2D Accuracy

| Model | overall |
|:----------------------------|----------:|
| Monkey-Chat | 72.6 |
| GPT-4v (detail: low) | 71.3 |
| Qwen-VL-Chat | 68.5 |
| Monkey | 67.6 |
| GeminiProVision | 66.7 |
| QwenVLPlus | 63.7 |
| Qwen-VL | 63.4 |
| LLaVA-InternLM2-20B (QLoRA) | 61.4 |
| CogVLM-17B-Chat | 60.3 |
| ShareGPT4V-13B | 59.3 |
| TransCore-M | 59.2 |
| LLaVA-v1.5-13B (QLoRA) | 59 |
| LLaVA-v1.5-13B | 57.9 |
| ShareGPT4V-7B | 56.7 |
| InternLM-XComposer-VL | 56.1 |
| LLaVA-InternLM-7B (QLoRA) | 56 |
| LLaVA-v1.5-7B (QLoRA) | 55.2 |
| mPLUG-Owl2 | 55.2 |
| SharedCaptioner | 55.1 |
| IDEFICS-80B-Instruct | 54.4 |
| LLaVA-v1.5-7B | 54.1 |
| PandaGPT-13B | 49.2 |
| LLaVA-v1-7B | 47.8 |
| IDEFICS-9B-Instruct | 42.7 |
| InstructBLIP-7B | 40.2 |
| VisualGLM | 40.2 |
| InstructBLIP-13B | 38.6 |
| MiniGPT-4-v1-13B | 33.4 |
| OpenFlamingo v2 | 30.7 |
| MiniGPT-4-v2 | 29.4 |
| MiniGPT-4-v1-7B | 28.7 |
58 changes: 30 additions & 28 deletions results/Caption.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,34 +10,36 @@

### Evaluation Results

| Model | BLEU-4 | BLEU-1 | ROUGE-L | CIDEr | Word_cnt mean. | Word_cnt std. |
|:------------------------------|---------:|---------:|----------:|--------:|-----------------:|----------------:|
| Qwen-VL-Chat | 34 | 75.8 | 54.9 | 98.9 | 10 | 1.7 |
| IDEFICS-80B-Instruct | 32.5 | 76.1 | 54.1 | 94.9 | 9.7 | 3.2 |
| IDEFICS-9B-Instruct | 29.4 | 72.7 | 53.4 | 90.4 | 10.5 | 4.4 |
| InstructBLIP-7B | 20.9 | 56.8 | 39.9 | 58.1 | 11.6 | 5.9 |
| InstructBLIP-13B | 16.9 | 50 | 37 | 52.4 | 11.8 | 12.8 |
| InternLM-XComposer-VL | 12.4 | 38.3 | 37.9 | 41 | 26.3 | 22.2 |
| TransCore-M | 8.8 | 30.3 | 36.1 | 34.7 | 39.9 | 27.9 |
| GeminiProVision | 8.4 | 33.2 | 31.2 | 9.7 | 35.2 | 15.7 |
| LLaVA-v1.5-7B (QLoRA, XTuner) | 7.2 | 25 | 36.6 | 43.2 | 48.8 | 42.9 |
| mPLUG-Owl2 | 7.1 | 25.8 | 33.6 | 35 | 45.8 | 32.1 |
| LLaVA-v1-7B | 6.7 | 27.3 | 26.7 | 6.1 | 40.9 | 16.1 |
| VisualGLM | 5.4 | 28.6 | 23.6 | 0.2 | 41.5 | 11.5 |
| LLaVA-v1.5-13B (QLoRA, XTuner) | 5.3 | 19.6 | 25.8 | 17.8 | 72.2 | 39.4 |
| LLaVA-v1.5-13B | 5.1 | 20.7 | 21.2 | 0.3 | 70.6 | 22.3 |
| LLaVA-v1.5-7B | 4.6 | 19.6 | 19.9 | 0.1 | 72.5 | 21.7 |
| PandaGPT-13B | 4.6 | 19.9 | 19.3 | 0.1 | 65.4 | 16.6 |
| MiniGPT-4-v1-13B | 4.4 | 20 | 19.8 | 1.3 | 64.4 | 30.5 |
| MiniGPT-4-v1-7B | 4.3 | 19.6 | 17.5 | 0.8 | 61.9 | 30.6 |
| LLaVA-InternLM-7B (QLoRA) | 4 | 17.3 | 17.2 | 0.1 | 82.3 | 21 |
| CogVLM-17B-Chat | 3.6 | 21.3 | 20 | 0.1 | 56.2 | 13.7 |
| Qwen-VL | 3.5 | 11.6 | 30 | 41.1 | 46.6 | 105.2 |
| GPT-4v (detail: low) | 3.3 | 18 | 18.1 | 0 | 77.8 | 20.4 |
| ShareGPT4V-7B | 1.4 | 9.7 | 10.6 | 0.1 | 147.9 | 45.4 |
| MiniGPT-4-v2 | 1.4 | 12.6 | 13.3 | 0.1 | 83 | 27.1 |
| OpenFlamingo v2 | 1.3 | 6.4 | 15.8 | 14.9 | 60 | 81.9 |
| SharedCaptioner | 1 | 8.8 | 9.2 | 0 | 164.2 | 31.6 |
| Model | BLEU-4 | BLEU-1 | ROUGE-L | CIDEr | Word_cnt mean. | Word_cnt std. |
|:----------------------------|---------:|---------:|----------:|--------:|-----------------:|----------------:|
| EMU2-Chat | 38.7 | 78.2 | 56.9 | 109.2 | 9.6 | 1.1 |
| Qwen-VL-Chat | 34 | 75.8 | 54.9 | 98.9 | 10 | 1.7 |
| IDEFICS-80B-Instruct | 32.5 | 76.1 | 54.1 | 94.9 | 9.7 | 3.2 |
| IDEFICS-9B-Instruct | 29.4 | 72.7 | 53.4 | 90.4 | 10.5 | 4.4 |
| InstructBLIP-7B | 20.9 | 56.8 | 39.9 | 58.1 | 11.6 | 5.9 |
| InstructBLIP-13B | 16.9 | 50 | 37 | 52.4 | 11.8 | 12.8 |
| InternLM-XComposer-VL | 12.4 | 38.3 | 37.9 | 41 | 26.3 | 22.2 |
| GeminiProVision | 8.4 | 33.2 | 31.2 | 9.7 | 35.2 | 15.7 |
| LLaVA-v1.5-7B (QLoRA) | 7.2 | 25 | 36.6 | 43.2 | 48.8 | 42.9 |
| mPLUG-Owl2 | 7.1 | 25.8 | 33.6 | 35 | 45.8 | 32.1 |
| LLaVA-v1-7B | 6.7 | 27.3 | 26.7 | 6.1 | 40.9 | 16.1 |
| VisualGLM | 5.4 | 28.6 | 23.6 | 0.2 | 41.5 | 11.5 |
| LLaVA-v1.5-13B (QLoRA) | 5.3 | 19.6 | 25.8 | 17.8 | 72.2 | 39.4 |
| LLaVA-v1.5-13B | 5.1 | 20.7 | 21.2 | 0.3 | 70.6 | 22.3 |
| LLaVA-v1.5-7B | 4.6 | 19.6 | 19.9 | 0.1 | 72.5 | 21.7 |
| PandaGPT-13B | 4.6 | 19.9 | 19.3 | 0.1 | 65.4 | 16.6 |
| MiniGPT-4-v1-13B | 4.4 | 20 | 19.8 | 1.3 | 64.4 | 30.5 |
| MiniGPT-4-v1-7B | 4.3 | 19.6 | 17.5 | 0.8 | 61.9 | 30.6 |
| LLaVA-InternLM-7B (QLoRA) | 4 | 17.3 | 17.2 | 0.1 | 82.3 | 21 |
| LLaVA-InternLM2-20B (QLoRA) | 4 | 17.9 | 17.3 | 0 | 83.2 | 20.4 |
| CogVLM-17B-Chat | 3.6 | 21.3 | 20 | 0.1 | 56.2 | 13.7 |
| Qwen-VL | 3.5 | 11.6 | 30 | 41.1 | 46.6 | 105.2 |
| GPT-4v (detail: low) | 3.3 | 18 | 18.1 | 0 | 77.8 | 20.4 |
| TransCore-M | 2.1 | 14.2 | 13.8 | 0.2 | 92 | 6.7 |
| ShareGPT4V-7B | 1.4 | 9.7 | 10.6 | 0.1 | 147.9 | 45.4 |
| MiniGPT-4-v2 | 1.4 | 12.6 | 13.3 | 0.1 | 83 | 27.1 |
| OpenFlamingo v2 | 1.3 | 6.4 | 15.8 | 14.9 | 60 | 81.9 |
| SharedCaptioner | 1 | 8.8 | 9.2 | 0 | 164.2 | 31.6 |

We noticed that, VLMs that generate long image descriptions tend to achieve inferior scores under different caption metrics.

Expand Down
61 changes: 33 additions & 28 deletions results/HallusionBench.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,31 +29,36 @@
> Models are sorted by the **descending order of qAcc.**


| Model | aAcc | fAcc | qAcc |
|:------------------------------|-------:|-------:|-------:|
| GPT-4v (detail: low) | 65.8 | 38.4 | 35.2 |
| GeminiProVision | 63.9 | 37.3 | 34.3 |
| Qwen-VL-Chat | 56.4 | 27.7 | 26.4 |
| MiniGPT-4-v1-7B | 52.4 | 17.3 | 25.9 |
| CogVLM-17B-Chat | 55.1 | 26.3 | 24.8 |
| InternLM-XComposer-VL | 57 | 26.3 | 24.6 |
| MiniGPT-4-v1-13B | 51.3 | 16.2 | 24.6 |
| SharedCaptioner | 55.6 | 22.8 | 24.2 |
| MiniGPT-4-v2 | 52.6 | 16.5 | 21.1 |
| InstructBLIP-7B | 53.6 | 20.2 | 19.8 |
| Qwen-VL | 57.6 | 12.4 | 19.6 |
| OpenFlamingo v2 | 52.7 | 17.6 | 18 |
| mPLUG-Owl2 | 48.9 | 22.5 | 16.7 |
| VisualGLM | 47.2 | 11.3 | 16.5 |
| IDEFICS-9B-Instruct | 50.1 | 16.2 | 15.6 |
| ShareGPT4V-7B | 48.2 | 21.7 | 15.6 |
| LLaVA-InternLM-7B (QLoRA) | 49.1 | 22.3 | 15.4 |
| InstructBLIP-13B | 47.9 | 17.3 | 15.2 |
| LLaVA-v1.5-7B | 48.3 | 19.9 | 14.1 |
| LLaVA-v1.5-13B (QLoRA, XTuner) | 46.9 | 17.6 | 14.1 |
| LLaVA-v1.5-7B (QLoRA, XTuner) | 46.2 | 16.2 | 13.2 |
| LLaVA-v1.5-13B | 46.7 | 17.3 | 13 |
| IDEFICS-80B-Instruct | 46.1 | 13.3 | 11 |
| TransCore-M | 44.7 | 16.5 | 10.1 |
| LLaVA-v1-7B | 44.1 | 13.6 | 9.5 |
| PandaGPT-13B | 43.1 | 9.2 | 7.7 |
| Model | aAcc | fAcc | qAcc |
|:----------------------------|-------:|-------:|-------:|
| GPT-4v (detail: low) | 65.8 | 38.4 | 35.2 |
| GeminiProVision | 63.9 | 37.3 | 34.3 |
| Monkey-Chat | 58.4 | 30.6 | 29 |
| Qwen-VL-Chat | 56.4 | 27.7 | 26.4 |
| MiniGPT-4-v1-7B | 52.4 | 17.3 | 25.9 |
| Monkey | 55.1 | 24 | 25.5 |
| CogVLM-17B-Chat | 55.1 | 26.3 | 24.8 |
| MiniGPT-4-v1-13B | 51.3 | 16.2 | 24.6 |
| InternLM-XComposer-VL | 57 | 26.3 | 24.6 |
| SharedCaptioner | 55.6 | 22.8 | 24.2 |
| MiniGPT-4-v2 | 52.6 | 16.5 | 21.1 |
| InstructBLIP-7B | 53.6 | 20.2 | 19.8 |
| Qwen-VL | 57.6 | 12.4 | 19.6 |
| OpenFlamingo v2 | 52.7 | 17.6 | 18 |
| EMU2-Chat | 49.4 | 22.3 | 16.9 |
| mPLUG-Owl2 | 48.9 | 22.5 | 16.7 |
| ShareGPT4V-13B | 49.8 | 21.7 | 16.7 |
| VisualGLM | 47.2 | 11.3 | 16.5 |
| TransCore-M | 49.7 | 21.4 | 15.8 |
| IDEFICS-9B-Instruct | 50.1 | 16.2 | 15.6 |
| ShareGPT4V-7B | 48.2 | 21.7 | 15.6 |
| LLaVA-InternLM-7B (QLoRA) | 49.1 | 22.3 | 15.4 |
| InstructBLIP-13B | 47.9 | 17.3 | 15.2 |
| LLaVA-InternLM2-20B (QLoRA) | 47.7 | 17.1 | 14.3 |
| LLaVA-v1.5-13B (QLoRA) | 46.9 | 17.6 | 14.1 |
| LLaVA-v1.5-7B | 48.3 | 19.9 | 14.1 |
| LLaVA-v1.5-7B (QLoRA) | 46.2 | 16.2 | 13.2 |
| LLaVA-v1.5-13B | 46.7 | 17.3 | 13 |
| IDEFICS-80B-Instruct | 46.1 | 13.3 | 11 |
| LLaVA-v1-7B | 44.1 | 13.6 | 9.5 |
| PandaGPT-13B | 43.1 | 9.2 | 7.7 |
Loading