Update inference_fast_vc3.py

inference_fast_vc3.py

@@ -311,20 +311,28 @@ class StreamingSession:
         round_idx: int,
         include_question: bool,
         frame_max_pixels: Optional[int] = None,
+        time_start: Optional[float] = None,
+        time_end: Optional[float] = None,
     ) -> List[dict]:
         """Build user message content for a single frame."""
-        time_tag = f"<{round_idx}s-{round_idx + 1}s>"
+        if time_start is None:
+            time_start = float(round_idx)
+        if time_end is None:
+            time_end = time_start + 1.0
+        time_tag = f"<{time_start:g}s-{time_end:g}s>"
         text = time_tag
         if include_question and self.question:
             text = f"{self.question}\n{text}"
-        image_item: dict = {"type": "image", "image": frame}
-        if frame_max_pixels is not None:
-            image_item["min_pixels"] = 28 * 28
-            image_item["max_pixels"] = frame_max_pixels
-        return [
-            image_item,
-            {"type": "text", "text": text},
-        ]
+        frames = frame if isinstance(frame, (list, tuple)) else [frame]
+        content = []
+        for one_frame in frames:
+            image_item: dict = {"type": "image", "image": one_frame}
+            if frame_max_pixels is not None:
+                image_item["min_pixels"] = 28 * 28
+                image_item["max_pixels"] = frame_max_pixels
+            content.append(image_item)
+        content.append({"type": "text", "text": text})
+        return content
 
     def _inject_question(self, msg: dict) -> None:
         """Prepend the question into the text part of a user message (in-place)."""
@@ -350,6 +358,8 @@ class StreamingSession:
         frame: Image.Image,
         round_idx: int,
         frame_max_pixels: Optional[int] = None,
+        time_start: Optional[float] = None,
+        time_end: Optional[float] = None,
     ) -> None:
         if not self._messages:
             # First frame: include question if global_question or round matches question_time.
@@ -363,7 +373,8 @@ class StreamingSession:
                 {
                     "role": "user",
                     "content": self._user_content(
-                        frame, round_idx, include_q, frame_max_pixels
+                        frame, round_idx, include_q, frame_max_pixels,
+                        time_start, time_end,
                     ),
                 }
             )
@@ -384,7 +395,8 @@ class StreamingSession:
             {
                 "role": "user",
                 "content": self._user_content(
-                    frame, round_idx, include_q, frame_max_pixels
+                    frame, round_idx, include_q, frame_max_pixels,
+                    time_start, time_end,
                 ),
             }
         )
@@ -430,9 +442,17 @@ class StreamingSession:
         frame: Image.Image,
         round_idx: int,
         frame_max_pixels: Optional[int] = None,
+        time_start: Optional[float] = None,
+        time_end: Optional[float] = None,
     ) -> str:
         """Feed one frame and return the model answer for this round."""
-        self._append_turn(frame, round_idx, frame_max_pixels=frame_max_pixels)
+        self._append_turn(
+            frame,
+            round_idx,
+            frame_max_pixels=frame_max_pixels,
+            time_start=time_start,
+            time_end=time_end,
+        )
 
         if self.debug:
             msgs = self._messages