Cache examples

app.py

@@ -76,7 +76,7 @@ def decode(container, sampling_rate, num_frames, clip_idx, num_clips, target_fps
 
 
 def get_video_duration(filename):
-    cap = cv2.VideoCapture(filename)
+    cap = cv2.VideoCapture(_extract_video_filepath(filename))
     if cap.isOpened():
         rate = cap.get(5)
         frame_num = cap.get(7)
@@ -85,13 +85,21 @@ def get_video_duration(filename):
     return -1
 
 
+def _extract_video_filepath(video_filename):
+    if isinstance(video_filename, dict):
+        return video_filename['video']['path']
+    return video_filename
+
+
 def predict_durations(model_checkpoint, text, video_filename, device="cpu"):
     print(f"Loading model: {model_checkpoint}")
     model = TvpForVideoGrounding.from_pretrained(model_checkpoint)
     processor = AutoProcessor.from_pretrained(model_checkpoint)
     print(f"Loading video: {video_filename}")
+    filepath = video_filename['video']['path'] if isinstance(video_filename, dict) else video_filename
     raw_sampled_frames = decode(
-        container=av.open(video_filename, metadata_errors="ignore"),
+        container=av.open(_extract_video_filepath(video_filename), metadata_errors="ignore"),
+        # container=av.open(video_filename['path'], metadata_errors="ignore"),
         sampling_rate=1,
         num_frames=model.config.num_frames,
         clip_idx=0,
@@ -114,15 +122,16 @@ HF_TOKEN = os.environ.get("HF_TOKEN", None)
 DEVICE = torch.device('cuda:0') if torch.cuda.is_available() else torch.device('cpu')
 MODELS = ["Intel/tvp-base",  "Intel/tvp-base-ANet"]
 EXAMPLES = [
-    ["./examples/bed.mp4", "a person is sitting on a bed."],
-    ["./examples/food.mp4", "a person eats some food."],
-    ["./examples/book.mp4", "a person reads a book."],
+    ["Intel/tvp-base", "a person is sitting on a bed.", "./examples/bed.mp4", ],
+    ["Intel/tvp-base", "a person eats some food.", "./examples/food.mp4", ],
+    ["Intel/tvp-base", "a person reads a book.", "./examples/book.mp4", ],
 ]
 
 model_checkpoint = gr.Dropdown(MODELS, label="Model", value=MODELS[0], type="value")
 video_in = gr.Video(label="Video File", elem_id="video_in")
-text_in = gr.Textbox(label="Text", placeholder="Description of event in the video", interactive=True)
-text_out = gr.Textbox(label="Prediction", placeholder="Predicted start and end time")
+# text_in = gr.Textbox(label="Text", placeholder="Description of event in the video", interactive=True)
+# text_out = gr.Textbox(label="Prediction", placeholder="Predicted start and end time")
+# examples = gr.Examples(examples=EXAMPLES, fn=predict_durations, inputs=[model_checkpoint, text_in, video_in], outputs=[text_out], cache_examples=True, preprocess=False)
 
 
 title = "Video Grounding with TVP"
@@ -131,20 +140,25 @@ css = """.toast-wrap { display: none !important } """
 with gr.Blocks(title=title) as demo:
     gr.Markdown(DESCRIPTION)
     with gr.Row():
-        model_checkpoint.render()
-
-    with gr.Row():
-        examples = gr.Examples(examples=EXAMPLES, inputs=[video_in, text_in])
+        model_checkpoint = gr.Dropdown(MODELS, label="Model", value=MODELS[0], type="value")
+        # model_checkpoint.render()
 
     with gr.Row():
         with gr.Column():
             video_in.render()
 
         with gr.Column():
-            text_in.render()
+            text_in = gr.Textbox(label="Text", placeholder="Description of event in the video", interactive=True)
+            text_out = gr.Textbox(label="Prediction", placeholder="Predicted start and end time")
+            # text_in #.render()
             time_button = gr.Button("Get start and end time")
             time_button.click(predict_durations, inputs=[model_checkpoint, text_in, video_in], outputs=[text_out])
-            text_out.render()
+            # text_out #.render()
+
+    with gr.Row():
+        examples = gr.Examples(examples=EXAMPLES, fn=predict_durations, inputs=[model_checkpoint, text_in, video_in], outputs=[text_out], cache_examples=True, preprocess=False)
+        # examples.render()
+        # text_out.render()
 
 
 if __name__ == "__main__":

gradio_cached_examples/15/log.csv

@@ -0,0 +1,4 @@
+Prediction,flag,username,timestamp
+"start: 0.0s, end: 6.8s",,,2023-11-22 15:49:43.930614
+"start: 0.0s, end: 11.4s",,,2023-11-22 15:50:00.348291
+"start: 0.0s, end: 5.6s",,,2023-11-22 15:50:16.641556