Add benchmark README
Browse files
README.md
ADDED
|
@@ -0,0 +1,119 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
pretty_name: FreeStyle Bench
|
| 3 |
+
task_categories:
|
| 4 |
+
- image-to-image
|
| 5 |
+
- text-to-image
|
| 6 |
+
language:
|
| 7 |
+
- en
|
| 8 |
+
tags:
|
| 9 |
+
- benchmark
|
| 10 |
+
- style-transfer
|
| 11 |
+
- image-editing
|
| 12 |
+
- cref
|
| 13 |
+
- sref
|
| 14 |
+
---
|
| 15 |
+
|
| 16 |
+
# FreeStyle Bench
|
| 17 |
+
|
| 18 |
+
This repository contains benchmark inputs for evaluating FreeStyle reference-based image editing.
|
| 19 |
+
|
| 20 |
+
It includes two benchmark subsets:
|
| 21 |
+
|
| 22 |
+
- `sref_bench/`: a style-transfer benchmark.
|
| 23 |
+
- `cref_sref_bench/`: a content-and-style dual-reference benchmark.
|
| 24 |
+
|
| 25 |
+
Both subsets use the same basic layout: a content-reference folder, a style-reference folder, and a `prompts.json` file that links each pair to its text prompt.
|
| 26 |
+
|
| 27 |
+
## Repository Layout
|
| 28 |
+
|
| 29 |
+
```text
|
| 30 |
+
<repo-root>/
|
| 31 |
+
README.md
|
| 32 |
+
sref_bench/
|
| 33 |
+
cref/
|
| 34 |
+
... content reference images ...
|
| 35 |
+
sref/
|
| 36 |
+
... style reference images ...
|
| 37 |
+
prompts.json
|
| 38 |
+
cref_sref_bench/
|
| 39 |
+
cref/
|
| 40 |
+
... content reference images ...
|
| 41 |
+
sref/
|
| 42 |
+
... style reference images ...
|
| 43 |
+
prompts.json
|
| 44 |
+
```
|
| 45 |
+
|
| 46 |
+
## `sref_bench`
|
| 47 |
+
|
| 48 |
+
`sref_bench` is the style-transfer benchmark.
|
| 49 |
+
|
| 50 |
+
For each benchmark pair:
|
| 51 |
+
|
| 52 |
+
- `sref_bench/cref/` contains the content reference image.
|
| 53 |
+
- `sref_bench/sref/` contains the style reference image.
|
| 54 |
+
- `sref_bench/prompts.json` contains the text prompt or instruction for the pair.
|
| 55 |
+
|
| 56 |
+
The keys in `prompts.json` are pair basenames. A key can be used to match the corresponding content and style reference images across the two folders.
|
| 57 |
+
|
| 58 |
+
For example, if `prompts.json` contains a key:
|
| 59 |
+
|
| 60 |
+
```text
|
| 61 |
+
example_pair
|
| 62 |
+
```
|
| 63 |
+
|
| 64 |
+
then the corresponding files should be found by matching the same basename under:
|
| 65 |
+
|
| 66 |
+
```text
|
| 67 |
+
sref_bench/cref/example_pair.*
|
| 68 |
+
sref_bench/sref/example_pair.*
|
| 69 |
+
```
|
| 70 |
+
|
| 71 |
+
where the image extension may depend on the actual uploaded files.
|
| 72 |
+
|
| 73 |
+
## `cref_sref_bench`
|
| 74 |
+
|
| 75 |
+
`cref_sref_bench` is the content-and-style dual-reference benchmark.
|
| 76 |
+
|
| 77 |
+
It has the same structure as `sref_bench`:
|
| 78 |
+
|
| 79 |
+
- `cref_sref_bench/cref/` contains the content reference image.
|
| 80 |
+
- `cref_sref_bench/sref/` contains the style reference image.
|
| 81 |
+
- `cref_sref_bench/prompts.json` contains the text prompt or instruction for the pair.
|
| 82 |
+
|
| 83 |
+
Again, each key in `prompts.json` is a pair basename and can be used to align the content reference image in `cref/` with the style reference image in `sref/`.
|
| 84 |
+
|
| 85 |
+
## How To Read A Pair
|
| 86 |
+
|
| 87 |
+
```python
|
| 88 |
+
import json
|
| 89 |
+
from pathlib import Path
|
| 90 |
+
from PIL import Image
|
| 91 |
+
|
| 92 |
+
bench_dir = Path("/path/to/FreeStyle_Bench/sref_bench") # or cref_sref_bench
|
| 93 |
+
|
| 94 |
+
with open(bench_dir / "prompts.json", "r", encoding="utf-8") as f:
|
| 95 |
+
prompts = json.load(f)
|
| 96 |
+
|
| 97 |
+
pair_name, prompt = next(iter(prompts.items()))
|
| 98 |
+
|
| 99 |
+
# Find images by matching the pair basename. The actual extension may vary.
|
| 100 |
+
cref_path = next((bench_dir / "cref").glob(pair_name + ".*"))
|
| 101 |
+
sref_path = next((bench_dir / "sref").glob(pair_name + ".*"))
|
| 102 |
+
|
| 103 |
+
content_ref = Image.open(cref_path).convert("RGB")
|
| 104 |
+
style_ref = Image.open(sref_path).convert("RGB")
|
| 105 |
+
|
| 106 |
+
print("pair:", pair_name)
|
| 107 |
+
print("prompt:", prompt)
|
| 108 |
+
print("content reference:", cref_path)
|
| 109 |
+
print("style reference:", sref_path)
|
| 110 |
+
```
|
| 111 |
+
|
| 112 |
+
## Notes
|
| 113 |
+
|
| 114 |
+
- `cref` means content reference.
|
| 115 |
+
- `sref` means style reference.
|
| 116 |
+
- `prompts.json` is the index file for each subset.
|
| 117 |
+
- The key of each `prompts.json` entry is the pair basename used to match images in `cref/` and `sref/`.
|
| 118 |
+
- Use `sref_bench` for style-transfer evaluation.
|
| 119 |
+
- Use `cref_sref_bench` for dual-reference content-and-style evaluation.
|