- [MINOR] [SOURCE] [UPDATE] 1. update app.py

app.py

@@ -9,6 +9,7 @@ import decord
 from decord import VideoReader
 import numpy as np
 import yaml
+import matplotlib.pyplot as plt
 
 from cover.datasets import UnifiedFrameSampler, spatial_temporal_view_decomposition
 from cover.models import COVER
@@ -25,6 +26,12 @@ mean_clip, std_clip = (
 
 sample_interval = 30
 
+comparison_array = {
+    "semantic": [3.0, 3.5, 2.5, 4.0, 2.0],  # 示例数组
+    "technical": [2.0, 3.0, 3.5, 4.0, 1.5],
+    "aesthetic": [2.5, 3.0, 2.0, 4.5, 3.5]
+}
+
 def get_sampler_params(video_path):
     vr = VideoReader(video_path)
     total_frames = len(vr)
@@ -44,13 +51,44 @@ def fuse_results(results: list):
         "overall"  : x,
     }
 
+def normalize_score(score, min_score=0, max_score=5):
+    return (score - min_score) / (max_score - min_score) * 5
+
+def compare_score(score, score_list):
+    better_than = sum(1 for s in score_list if score > s)
+    percentage = better_than / len(score_list) * 100
+    return f"Better than {percentage:.0f}% videos in YT-UGC" if percentage > 50 else f"Worse than {100-percentage:.0f}% videos in YT-UGC"
+
+def create_bar_chart(scores, comparisons):
+    labels = ['Semantic', 'Technical', 'Aesthetic', 'Overall']
+    colors = ['#d62728', '#1f77b4', '#ff7f0e', '#bcbd22']
+    
+    fig, ax = plt.subplots(figsize=(10, 5))
+    
+    for i, (label, score, comparison, color) in enumerate(zip(labels, scores, comparisons, colors)):
+        ax.barh(i, score, color=color, edgecolor='black')
+        ax.text(score, i, f'{score:.1f}', va='center', ha='left')
+        ax.text(5.1, i, comparison, va='center', ha='left')
+    
+    ax.set_yticks(range(len(labels)))
+    ax.set_yticklabels(labels)
+    ax.set_xlim(0, 5)
+    ax.set_xlabel('Score')
+    
+    plt.tight_layout()
+    image_path = "./bar_chart.png"
+    plt.savefig(image_path)
+    plt.close()
+    
+    return image_path
+
 def inference_one_video(input_video):
     """
     BASIC SETTINGS
     """
     device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
     with open("./cover.yml", "r") as f:
-       opt = yaml.safe_load(f)
+        opt = yaml.safe_load(f)
     
     dopt = opt["data"]["val-ytugc"]["args"]
     temporal_samplers = {}
@@ -112,14 +150,30 @@ def inference_one_video(input_video):
 
     results = [r.mean().item() for r in evaluator(views)]
     pred_score = fuse_results(results)
-    return pred_score
+    
+    normalized_scores = [
+        normalize_score(pred_score["semantic"]),
+        normalize_score(pred_score["technical"]),
+        normalize_score(pred_score["aesthetic"]),
+        normalize_score(pred_score["overall"])
+    ]
+    
+    comparisons = [
+        compare_score(pred_score["semantic"], comparison_array["semantic"]),
+        compare_score(pred_score["technical"], comparison_array["technical"]),
+        compare_score(pred_score["aesthetic"], comparison_array["aesthetic"]),
+        compare_score(pred_score["overall"], comparison_array["semantic"] + comparison_array["technical"] + comparison_array["aesthetic"])  # 假设 overall 分数的比较使用所有维度分数的组合
+    ]
+    
+    image_path = create_bar_chart(normalized_scores, comparisons)
+    return image_path
 
 # Define the input and output types for Gradio using the new API
 video_input = gr.Video(label="Input Video")
-output_label = gr.JSON(label="Scores")
+output_image = gr.Image(label="Scores")
 
 # Create the Gradio interface
-gradio_app = gr.Interface(fn=inference_one_video, inputs=video_input, outputs=output_label)
+gradio_app = gr.Interface(fn=inference_one_video, inputs=video_input, outputs=output_image)
 
 if __name__ == "__main__":
     gradio_app.launch()