fixing some stuff

requirements.txt

@@ -13,3 +13,4 @@ together
 einops
 opencv-python
 timm
+#inference

src/labeler.py

@@ -16,9 +16,6 @@ class TogetherLLMLabeler:
         Commentary:
         {transcript}
 
-        Spatial Context (object detections per frame):
-        {spatial_context}
-
         Instructions:
         - Summarize this event in factual soccer terminology.
         - Focus on the play's significance to the score.
@@ -35,4 +32,8 @@ class TogetherLLMLabeler:
             max_tokens=200
         )
 
-        return response.choices[0].message["content"].strip()
+        return response.choices[0].message["content"].strip()
+    
+#after commentary:
+#        Spatial Context (object detections per frame):
+#        {spatial_context}

src/segmenter.py

@@ -1,51 +1,40 @@
 import cv2
-import os
-from roboflow import Roboflow
-from dotenv import load_dotenv
-
-load_dotenv() 
-## When the ball is no longer detected, we start a new segment
-
-def detect_event_segments(video_path, confidence=0.4):
-    rf = Roboflow(api_key=os.getenv("ROBOFLOW_API_KEY"))
-    project = rf.workspace().project("soccer-players-ckbru/15")
-    model = project.version(1).model
 
+def detect_event_segments(video_path):
     cap = cv2.VideoCapture(video_path)
     fps = cap.get(cv2.CAP_PROP_FPS)
 
-    events = []
-    active_event = None
-    frame_data = []
+    segments = []
+    segment_duration = 5  # seconds per segment
+    frames_per_segment = int(segment_duration * fps)
+
+    frame_number = 0
 
     while cap.isOpened():
         ret, frame = cap.read()
         if not ret:
             break
 
-        frame_number = int(cap.get(cv2.CAP_PROP_POS_FRAMES))
-        detections = model.predict(frame, confidence=confidence).json().get('predictions', [])
-        frame_data.append({"frame": frame_number, "objects": detections})
-
-        ball_detected = any(obj['class'] == 'ball' for obj in detections)
-        goal_area_activity = any(obj['class'] == 'goal' for obj in detections) and ball_detected
-
-        if goal_area_activity and active_event is None:
-            active_event = {"start_frame": frame_number, "frames": []}
+        frame_number += 1
 
-        if active_event:
-            active_event["frames"].append(frame_data[-1])
+        if frame_number % frames_per_segment == 1:
+            segment_start_sec = (frame_number - 1) / fps
+            segment_end_sec = (frame_number + frames_per_segment - 2) / fps
+            segments.append({
+                "start_sec": segment_start_sec,
+                "end_sec": segment_end_sec,
+                "frames": []  # This can hold keyframes later if needed
+            })
 
-        if active_event and not ball_detected:
-            active_event["end_frame"] = frame_number
-            events.append(active_event)
-            active_event = None
+        if segments:
+            segments[-1]["frames"].append(frame)
 
     cap.release()
 
-    # Convert frames to timestamps
-    for event in events:
-        event['start_sec'] = event['start_frame'] / fps
-        event['end_sec'] = event['end_frame'] / fps
+    # Final cleanup to make sure segment end matches actual video length if needed
+    if segments:
+        total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
+        total_duration = total_frames / fps
+        segments[-1]["end_sec"] = min(segments[-1]["end_sec"], total_duration)
 
-    return events
+    return segments

src/utils.py

@@ -35,7 +35,7 @@ def generate_frame_urls(frame_paths):
     base_url = os.getenv("SPACE_URL", "http://localhost:8000")
     return [f"{base_url}/data/{os.path.basename(path)}" for path in frame_paths]
 
-def match_transcript_to_events(events, transcript):
+def match_transcript_to_segments(events, transcript):
     for event in events:
         matched_lines = [
             line["text"] for line in transcript