Merge pull request #388 from modelscope/preference_model

Preference model

examples/image_quality_metric/README.md

@@ -0,0 +1,15 @@
+# Image Quality Metric
+
+The image quality assessment functionality has been integrated into Diffsynth. We support the following models:
+
+* [ImageReward](https://github.com/THUDM/ImageReward)
+* [Aesthetic](https://github.com/christophschuhmann/improved-aesthetic-predictor)
+* [PickScore](https://github.com/yuvalkirstain/pickscore)
+* [CLIP](https://github.com/openai/CLIP)
+* [HPSv2](https://github.com/tgxs002/HPSv2)
+* [HPSv2.1](https://github.com/tgxs002/HPSv2)
+* [MPS](https://github.com/Kwai-Kolors/MPS)
+
+## Usage
+
+See [`./image_quality_evaluation.py`](./image_quality_evaluation.py) for more details.

examples/image_quality_metric/image_quality_evaluation.py

@@ -0,0 +1,23 @@
+from diffsynth.extensions.ImageQualityMetric import download_preference_model, load_preference_model
+from modelscope import dataset_snapshot_download
+from PIL import Image
+
+
+# Download example image
+dataset_snapshot_download(
+    dataset_id="DiffSynth-Studio/examples_in_diffsynth",
+    allow_file_pattern="data/examples/ImageQualityMetric/image.jpg",
+    local_dir="./"
+)
+
+# Parameters
+prompt = "an orange cat"
+image = Image.open("data\examples\ImageQualityMetric\image.jpg")
+device = "cuda"
+cache_dir = "./models"
+
+# Run preference models
+for model_name in ["ImageReward", "Aesthetic", "PickScore", "CLIP", "HPSv2", "HPSv2.1", "MPS"]:
+    path = download_preference_model(model_name, cache_dir=cache_dir)
+    preference_model = load_preference_model(model_name, device=device, path=path)
+    print(model_name, preference_model.score(image, prompt))