zhangzicheng
commited on
Commit
•
64953d3
1
Parent(s):
9de52d7
Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,43 @@
|
|
1 |
---
|
2 |
license: mit
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: mit
|
3 |
---
|
4 |
+
|
5 |
+
## This is the benchmark dataset for ["A Benchmark for Multi-modal Foundation Models on Low-level Vision: from Single Images to Pairs"](https://arxiv.org/abs/2402.07116)
|
6 |
+
|
7 |
+
# The structure of the dataset is as follows:
|
8 |
+
|
9 |
+
1. q-bench2-a1-dev.jsonl (**with** *img_path*, *question*, *answer_candidates*, *correct_answer*)
|
10 |
+
2. q-bench2-a1-test.jsonl (**with** *img_path*, *question*, *answer_candidates*, **without** *correct_answer*)
|
11 |
+
3. q-bench2-a2.jsonl (**with** *img_path*, *empty response*)
|
12 |
+
4. q-bench-compare.zip (image files)
|
13 |
+
|
14 |
+
# The img_path is organized as *prefix* + *img1* + \_cat\_ + *img2* + *.jpg*
|
15 |
+
|
16 |
+
For example, if the img_path is "llvisionqa_compare_dev\\\\00079.jpg_cat_09769.jpg.jpg", then the prefix is "llvisionqa_compare_dev", img1 is "00079.jpg", img2 is "09769.jpg".
|
17 |
+
|
18 |
+
You can use the function to get the image paths:
|
19 |
+
|
20 |
+
```
|
21 |
+
def get_img_names(img_path, prefix = "path_to_all_single_images"):
|
22 |
+
img_paths = img_path.split('\\')[1][:-4].split("_cat_")
|
23 |
+
img1_name = os.path.join(prefix,img_paths[0])
|
24 |
+
img2_name = os.path.join(prefix,img_paths[1])
|
25 |
+
return img1_name,img2_name
|
26 |
+
```
|
27 |
+
# The q-bench-compare.zip structure is:
|
28 |
+
|
29 |
+
1. all_single_images: all of the single images used
|
30 |
+
2. llvisionqa_compare_dev: the concatenated images for the dev subset of the perception-compare task
|
31 |
+
3. llvisionqa_compare_test: the concatenated images for the test subset of the perception-compare task
|
32 |
+
4. lldescribe_compare: the concatenated images for the description-compare task
|
33 |
+
|
34 |
+
# Submission for test your own MLLM on q-bench2
|
35 |
+
|
36 |
+
1. Perception-compare task (a1): organize your jsonl file "q-bench2-a1-test_(YOUR_MLLM_NAME).jsonl" as the structure of the provided "q-bench2-a1-dev.jsonl"
|
37 |
+
2. Description-compare task (a2): simply complete the empty "response" of "q-bench2-a2.jsonl" file and rename into "q-bench2-a2_(YOUR_MLLM_NAME).jsonl"
|
38 |
+
|
39 |
+
Please contact any of the first authors to get the results of your MLLM with the submission files.
|
40 |
+
|
41 |
+
Zicheng Zhang, zzc1998@sjtu.edu.cn
|
42 |
+
Haoning Wu, haoning001@e.ntu.edu.sg
|
43 |
+
|