Address comment

README.md

@@ -127,8 +127,14 @@ from sam2.sam2_video_predictor import SAM2VideoPredictor
 predictor = SAM2VideoPredictor.from_pretrained("facebook/sam2-hiera-large")
 
 with torch.inference_mode(), torch.autocast("cuda", dtype=torch.bfloat16):
-    predictor.set_image(<your_image>)
-    masks, _, _ = predictor.predict(<input_prompts>)
+    state = predictor.init_state(<your_video>)
+
+    # add new prompts and instantly get the output on the same frame
+    frame_idx, object_ids, masks = predictor.add_new_points(state, <your_prompts>):
+
+    # propagate the prompts to get masklets throughout the video
+    for frame_idx, object_ids, masks in predictor.propagate_in_video(state):
+        ...
 ```
 
 ## Model Description