add benchmark results

Q-Future · Nov 6, 2023 · 721b8c3 · 721b8c3
1 parent 77b03a6
commit 721b8c3
Show file tree

Hide file tree

Showing 5 changed files with 41 additions and 0 deletions.
diff --git a/benchmark_results/README.md b/benchmark_results/README.md
@@ -0,0 +1,41 @@
+## Results
+
+### Answering Multi-Choice Questions
+
+#### Quantitative Results
+
+
+- `dev` subset of [LLVisionQA](https://huggingface.co/datasets/nanyangtu/LLVisionQA-QBench).
+
+  <div style="width: 100%; text-align: center; margin:auto;">
+      <img style="width:100%" src="mcq_dev.png">
+  </div> 
+
+- `test` subset of [LLVisionQA](https://huggingface.co/datasets/nanyangtu/LLVisionQA-QBench).
+
+  <div style="width: 100%; text-align: center; margin:auto;">
+      <img style="width:100%" src="mcq_test.png">
+  </div> 
+
+### General Description on Low-level Visual Aspects
+
+#### Quantitative Results
+
+- Results on [LLDescribe](https://huggingface.co/datasets/nanyangtu/LLDescribe-QBench).
+
+  <div style="width: 100%; text-align: center; margin:auto;">
+      <img style="width:100%" src="description.png">
+  </div> 
+
+
+### Image and Video Quality Assessment
+
+#### Quantitative Results
+
+- Impressive results on 8 IQA/VQA datasets, including 3 *never seen* datasets (CGIQA-6K, KADID-10K, KoNViD-10k).
+
+  <div style="width: 100%; text-align: center; margin:auto;">
+      <img style="width:100%" src="iqa_vqa.png">
+  </div> 
+
+- The results are obtained with text-only instruction tuning, **without any numerical supervision**.
diff --git a/benchmark_results/description.png b/benchmark_results/description.png
diff --git a/benchmark_results/iqa_vqa.png b/benchmark_results/iqa_vqa.png
diff --git a/benchmark_results/mcq_dev.png b/benchmark_results/mcq_dev.png
diff --git a/benchmark_results/mcq_test.png b/benchmark_results/mcq_test.png