infv1

docker-compose.yml

@@ -14,15 +14,15 @@ services:
     depends_on:
       - redis
     environment:
-#      stream_url: "rtsp://admin:labvision2019@10.1.8.182:554"
-      stream_url: "/videos/video.mp4"
+      stream_url: "rtsp://admin:labvision2019@10.1.8.182:554"
+#      stream_url: "/videos/video.mp4"
       stream_label: "pelea3"
       redis_host: "redis"
       redis_port: "6379"
       redis_db: "0"
       restart: unless-stopped
-    volumes:
-      - /home/ifiguero/DIA/dia3/DIA0205 - VISIÓN COMPUTACIONAL APLICADA/Profesor Cristian Aguilera/dataset/pelea3.mp4:/videos/video.mp4:ro
+    # volumes:
+    #   - /home/ifiguero/DIA/dia3/DIA0205 - VISIÓN COMPUTACIONAL APLICADA/Profesor Cristian Aguilera/dataset/pelea3.mp4:/videos/video.mp4:ro
 
   stream_preprocess:
     build:
@@ -32,6 +32,7 @@ services:
     depends_on:
       - redis
       - stream_reader
+      - stream_inference
     environment:
       stream_label: "pelea3"
       redis_host: "redis"
@@ -39,15 +40,45 @@ services:
       redis_db: "0"
       restart: unless-stopped
 
-  dataset_builder:
+  stream_inference:
     build:
       dockerfile: Dockerfile
-      context: dataset
-    container_name: dataset_builder
+      context: inference
+    container_name: stream_inference
+    # deploy:
+    #   resources:
+    #     reservations:
+    #       devices:
+    #         - driver: nvidia
+    #           count: all
+    #           capabilities: [gpu]
     depends_on:
       - redis
-      - stream_preprocess
     environment:
+      model_folder: "/app/models"
+      model_name: "tuned"
+      out_folder: "/app/out_folder"
+      stream_label: "pelea3"
+      threshold: "0.75"
+      redis_host: "redis"
+      redis_port: "6379"
+      redis_db: "0"
+      restart: unless-stopped
+    volumes:
+      - ./rt_out:/app/out_folder
+      - ./models:/app/models
+
+  web_inference:
+    build:
+      dockerfile: Dockerfile
+      context: web
+    container_name: web_inference
+    depends_on:
+      - redis
+      - stream_inference
+    environment:
+      stream_url: "rtsp://admin:labvision2019@10.1.8.182:554"
+#      stream_url: "/videos/video.mp4"
       out_folder: "/app/out_folder"
       stream_label: "pelea3"
       redis_host: "redis"
@@ -55,4 +86,6 @@ services:
       redis_db: "0"
       restart: unless-stopped
     volumes:
-      - ./out_folder:/app/out_folder
+      - ./rt_out:/app/out_folder
+    ports:
+      - "8080:8080"

inference/Dockerfile

@@ -26,10 +26,10 @@ RUN /opt/conda/bin/pip install --upgrade pip && \
     pillow
 
 RUN /opt/conda/bin/conda install -y -c conda-forge opencv
-RUN /opt/conda/bin/pip install --force-reinstall --no-cache-dir tqdm
+RUN /opt/conda/bin/pip install --force-reinstall --no-cache-dir redis 
 
 # Copy project files into container
-COPY evaluate.py /app/evaluate.py
+COPY inference.py /app/inference.py
 
 # Default command (you can override with docker run)
-CMD ["python", "/app/evaluate.py"]
+CMD ["python", "/app/inference.py"]

inference/inference.py

@@ -17,16 +17,19 @@ redis_host = os.getenv("redis_host", "localhost")
 redis_port = int(os.getenv("redis_port", 6379))
 stream_label = os.getenv("stream_label", "cropped_stream")
 output_folder = os.getenv("out_folder", "/app/out_folder")
-model_folder = os.getenv("model_folder", "base_model")
-model_path = os.path.join(output_folder, model_folder)
-threshold = float(os.getenv("threshold", "0.5"))
-queue_label = f"{stream_label}_cubes"
+model_folder = os.getenv("model_folder", "/app/models")
+model_name = os.getenv("model_name", "base_model")
 
+threshold = float(os.getenv("threshold", "0.5"))
+
+queue_label = f"{stream_label}_cubes"
+model_path = os.path.join(model_folder, model_name)
 sqlite_path = os.path.join(output_folder, f"{stream_label}_results.db")
 stream_folder = os.path.join(output_folder, stream_label)
 
 os.makedirs(output_folder, exist_ok=True)
 os.makedirs(stream_folder, exist_ok=True)
+os.makedirs(model_path, exist_ok=True)
 
 # -----------------------------
 # CONNECT TO REDIS
@@ -43,7 +46,7 @@ model.eval()
 
 # Warm up
 with torch.no_grad():
-    dummy = torch.randn(1, 3, 16, 224, 224).to(device)
+    dummy = torch.randn(1, 16, 3, 224, 224).to(device)
     _ = model(pixel_values=dummy)
 
 # -----------------------------
@@ -55,7 +58,7 @@ cursor.execute("""
 CREATE TABLE IF NOT EXISTS inference_results (
     id INTEGER PRIMARY KEY AUTOINCREMENT,
     gif_name TEXT,
-    timestamp TEXT,
+    timestamp TIMESTAMP,
     prediction INTEGER
 )
 """)
@@ -69,6 +72,7 @@ def from_redis_list(queue_label):
     while True:
         compressed_data = redis_conn.lpop(queue_label)
         if compressed_data:
+            retry = 0
             buffer = io.BytesIO(compressed_data)
             loaded_data = np.load(buffer)
             frames = [loaded_data[key] for key in sorted(loaded_data.files)]
@@ -78,49 +82,53 @@ def from_redis_list(queue_label):
             if retry % 50 == 0:
                 print(f"[WAIT] Queue {queue_label} empty for {retry/50:.1f} seconds...")
             time.sleep(1/50.0)
-            if retry > 2000:
-                raise TimeoutError(f"Queue {queue_label} empty for over 40s")
+            if retry > 3000:
+                raise TimeoutError(f"Queue {queue_label} empty for over 1m")
 
 # -----------------------------
 # SAVE GIF
 # -----------------------------
 def save_gif(frames, gif_path):
-    images = [Image.fromarray(frame.astype(np.uint8)) for frame in frames]
-    images[0].save(gif_path, save_all=True, append_images=images[1:], duration=50, loop=0)
+    images = [Image.fromarray(frame[:, :, ::-1]) for frame in frames]
+    images[0].save(gif_path, save_all=True, append_images=images[1:], duration=1, loop=0)
+    # images = [Image.fromarray(frame.astype(np.uint8)) for frame in frames]
+    # images[0].save(gif_path, save_all=True, append_images=images[1:], duration=50, loop=0)
 
 # -----------------------------
 # MAIN LOOP
 # -----------------------------
 print(f"[INFO] Listening on Redis queue: {queue_label}")
-while True:
-    try:
-        frames = from_redis_list(queue_label)
-        if not all(frame.shape == (224, 224, 3) for frame in frames):
-            print("[WARN] Skipped frame batch due to incorrect shape")
-            continue
+if __name__ == "__main__":
+    while True:
+        try:
+            frames = from_redis_list(queue_label)
+            # if not all(frame.shape == (224, 224, 3) for frame in frames):
+            #     print("[WARN] Skipped frame batch due to incorrect shape")
+            #     continue
 
-        timestamp = datetime.utcnow().strftime("%y%m%d_%H%M%S_%f")[:-3]
-        gif_filename = f"{stream_label}/{stream_label}_{timestamp}.gif"
-        gif_path = os.path.join(stream_folder, gif_filename)
+            now =  datetime.utcnow()
+            timestamp = now.strftime("%y%m%d_%H%M%S_%f")[:-3]
+            gif_filename = f"{stream_label}_{timestamp}.gif"
+            gif_path = os.path.join(stream_folder, gif_filename)
 
-        # Save GIF
-        save_gif(frames, gif_path)
+            # Save GIF
+            save_gif(frames, gif_path)
 
-        # Preprocess and predict
-        with torch.no_grad():
-            inputs = processor(images=[frame[:, :, ::-1] for frame in frames], return_tensors="pt")  # convert BGR to RGB
-            pixel_values = inputs["pixel_values"].to(device)  # [1, 3, 16, 224, 224]
-            logits = model(pixel_values=pixel_values).logits
-            prob = torch.softmax(logits, dim=-1)[0][1].item()
-            prediction = int(prob > threshold)
+            # Preprocess and predict
+            with torch.no_grad():
+                inputs = processor(images=[frame[:, :, ::-1] for frame in frames], return_tensors="pt")  # convert BGR to RGB
+                pixel_values = inputs["pixel_values"].to(device)  # [1, 3, 16, 224, 224]
+                logits = model(pixel_values=pixel_values).logits
+                prob = torch.softmax(logits, dim=-1)[0][1].item()
+                prediction = int(prob > threshold)
 
-        # Insert into SQLite
-        cursor.execute("INSERT INTO inference_results (gif_name, timestamp, prediction) VALUES (?, ?, ?)",
-                       (gif_filename, timestamp, prediction))
-        conn.commit()
+            # Insert into SQLite
+            cursor.execute("INSERT INTO inference_results (gif_name, timestamp, prediction) VALUES (?, ?, ?)",
+                           (gif_filename, now, prediction))
+            conn.commit()
 
-        print(f"[INFO] Saved {gif_filename} | Class={prediction} | Prob={prob:.3f}")
-
-    except Exception as e:
-        print(f"[ERROR] {e}")
-        time.sleep(1)
+            print(f"[INFO] Saved {gif_filename} | Class={prediction} | Prob={prob:.3f}")
+        except TimeoutError as e:
+            print(f"[TimeoutError] {e}")
+        except Exception as e:
+            print(f"[ERROR] {e}")

preprocess/preprocess.py

@@ -31,7 +31,7 @@ def fromRedis(queue_label):
         compressed_data = redis_conn.lpop(queue_label)
         if compressed_data:
             retry = 0
-            print(f"Popped data from queue: {queue_label}")
+#            print(f"Popped data from queue: {queue_label}")
             buffer = io.BytesIO(compressed_data)
             loaded_data = np.load(buffer)
             return loaded_data['data']
@@ -41,10 +41,10 @@ def fromRedis(queue_label):
                 print(f"Queue {queue_label} empty for {retry/50} seconds")
             time.sleep(1/50.0)
             if retry > 1000:
-                raise(f'Queue {queue_label} 20s empty')
+                raise TimeoutError(f'Queue {queue_label} 20s empty')
 
 def toRedisList(queue_label, data_list):
-    print(f"Pushed data to queue: {queue_label}")
+#    print(f"Pushed data to queue: {queue_label}")
     buffer = io.BytesIO()
     np.savez(buffer, *data_list)  # Use *data_list to unpack the list into arguments for savez_compressed
     compressed_data = buffer.getvalue()
@@ -57,7 +57,7 @@ def fromRedisList(queue_label):
         compressed_data = redis_conn.lpop(queue_label)
         if compressed_data:
             retry = 0
-            print(f"Popped data from queue: {queue_label}")
+#            print(f"Popped data from queue: {queue_label}")
             buffer = io.BytesIO(compressed_data)
             loaded_data = np.load(buffer)
             data_list = [loaded_data[key] for key in loaded_data.files]
@@ -69,7 +69,7 @@ def fromRedisList(queue_label):
             time.sleep(1/50.0)
 
             if retry > 1000:
-                raise(f'Queue {queue_label} 20s empty')
+                raise TimeoutError(f'Queue {queue_label} 20s empty')
 
 
 import mediapipe as mp
@@ -106,7 +106,7 @@ def detect_person_bbox(image_frame: np.ndarray) -> list:
                     width, height = bbox.width, bbox.height
 
                 padding = int(max(10, width//10))
-                
+
                 width  += 2 * padding
                 height += 2 * padding
                 left = max(0, left-padding)
@@ -163,6 +163,8 @@ def crop_and_resize_frames(frames, box, target_size=224):
         cropped.append(crop)
     return cropped
 
+import collections
+
 if __name__ == "__main__":
     frame_list = []
     frame_count = 0
@@ -173,33 +175,39 @@ if __name__ == "__main__":
     lap_time = start_time
 
     print(f"[INFO] Starting consumer for stream: {stream_label}")
-    while frame_lag < 100:
-        frame = fromRedis(stream_label)
-        if frame is None:
-            frame_lag += 1
-            wait_time = frame_lag * 0.01
-            time.sleep(wait_time)
-            continue
-        frame_lag = 0
+    try:
+        while True:
+            frame = fromRedis(stream_label)
 
-        frame_list.append(frame)
-        frame_count += 1
+            frame_list.append(frame)
+            frame_count += 1
 
-        if len(frame_list) == 16:
-            for box in  detect_person_bbox(frame_list[0]):
-                frame_hits += 1
-                cropped_frames = crop_and_resize_frames(frame_list, box)
-                toRedisList(stream_label_queue, cropped_frames)
+            if len(frame_list) == 16:
+                hits = frame_hits
+                for box in  detect_person_bbox(frame_list[0]):
+                    frame_hits += 1
+                    cropped_frames = crop_and_resize_frames(list(frame_list), box)
+                    toRedisList(stream_label_queue, cropped_frames)
 
-            frame_list.pop(0)
+                if frame_hits > hits:
+                    frame_list.clear()
+                else:
+                    frame_list.pop(0)
+                    frame_list.pop(0)
+                    frame_list.pop(0)
+                    frame_list.pop(0)
 
-        if frame_count % 15 == 0:
-            current_hits = frame_hits - last_hits
-            last_hits = frame_hits
-            last_lap = lap_time
-            lap_time = time.time()
-            elapsed =  lap_time - last_lap
-            print(f"[INFO] {frame_count} frames, {frame_hits} hits. {current_hits} in {elapsed:.2f} seconds.")
+            if frame_count % 15 == 0:
+                current_hits = frame_hits - last_hits
+                last_hits = frame_hits
+                last_lap = lap_time
+                lap_time = time.time()
+                elapsed =  lap_time - last_lap
+                print(f"[INFO] {frame_count} frames, {frame_hits} hits. {current_hits} in {elapsed:.2f} seconds.")
+    except TimeoutError as e:
+        print(f"[TimeoutError] {e}")
+    except Exception as e:
+        print(f"[ERROR] {e}")
 
     total_time = time.time() - start_time
     print(f"[INFO] Finished. Total frames: {frame_count}. Total Hits {frame_hits}. Total time: {total_time:.2f} seconds.")

reader/reader.py

@@ -23,7 +23,7 @@ redis_conn.delete(stream_label)
 print(f"[INFO] Cleared Redis queue for label: {stream_label}")
 
 def toRedis(queue_label, data):
-    print(f"Pushed data to queue: {queue_label}")
+#    print(f"Pushed data to queue: {queue_label}")
     buffer = io.BytesIO()
     np.savez(buffer, data=data)
     compressed_data = buffer.getvalue()
@@ -38,11 +38,9 @@ def fromRedis(queue_label):
             buffer = io.BytesIO(compressed_data)
             loaded_data = np.load(buffer)
             return loaded_data['data']
-        else:
-            time.sleep(1/50.0)
 
 def toRedisList(queue_label, data_list):
-    print(f"Pushed data to queue: {queue_label}")
+#    print(f"Pushed data to queue: {queue_label}")
     buffer = io.BytesIO()
     np.savez(buffer, *data_list)  # Use *data_list to unpack the list into arguments for savez_compressed
     compressed_data = buffer.getvalue()
@@ -58,8 +56,6 @@ def fromRedisList(queue_label):
             loaded_data = np.load(buffer)
             data_list = [loaded_data[key] for key in loaded_data.files]
             return data_list
-        else:
-            time.sleep(1/50.0)
 
 
 # Main frame processing function
@@ -86,14 +82,14 @@ def stream_frames(video_url, label):
 
         toRedis(label,frame)
         frame_count += 1
-        time.sleep(1/30.0)
+#        time.sleep(1/30.0)
 
         # Log every 500 frames
-        if frame_count % 100 == 0:
+        if frame_count % 16 == 0:
             last_lap = lap_time
             lap_time = time.time()
             elapsed =  lap_time - last_lap
-            print(f"[INFO] 100 frames processed in {elapsed:.2f} seconds.")
+            print(f"[INFO] 16 frames processed in {elapsed:.2f} seconds.")
 
         # if frame_count % 16 and (redis_len := redis_conn.llen(label)) > 75:
         #     print(f"[WARN] {redis_len} items in queue. Pruning to 50 latest.")

web/Dockerfile

@@ -0,0 +1,21 @@
+# Base Python image
+FROM python:3.10-slim
+
+# Set workdir
+WORKDIR /app
+
+
+# Install dependencies
+RUN pip install --no-cache-dir flask redis
+
+
+# Copy files into the container
+COPY templates /app/templates
+COPY web.py /app/web.py
+
+# Expose port for Flask app
+EXPOSE 8080
+
+
+# Run the Flask app
+CMD ["python", "web.py"]

web/templates/index.html

@@ -0,0 +1,126 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <title>Stream Monitor</title>
+    <meta name="viewport" content="width=device-width, initial-scale=1">
+
+    <!-- Bootstrap 5 -->
+    <link href="https://cdn.jsdelivr.net/npm/bootstrap@5.3.3/dist/css/bootstrap.min.css" rel="stylesheet">
+
+    <style>
+        body {
+            background-color: #f8f9fa;
+        }
+        .video-frame {
+            width: 100%;
+            aspect-ratio: 16 / 9;
+            background: black;
+        }
+        .gif-preview {
+            width: 100%;
+            border-radius: 5px;
+        }
+        .detection-list {
+            max-height: 80vh;
+            overflow-y: auto;
+        }
+        .section-title {
+            margin-top: 1rem;
+            margin-bottom: 1rem;
+            font-weight: bold;
+        }
+        .small-text {
+            font-size: 0.85rem;
+            color: #666;
+        }
+    </style>
+</head>
+<body>
+<div class="container-fluid mt-4">
+    <div class="row">
+
+        <!-- Live Stream -->
+        <div class="col-md-6">
+            <h4 class="section-title">Live Stream</h4>
+            <div class="video-frame">
+                <!-- Placeholder RTSP (replace with actual player integration) -->
+                <video controls autoplay muted width="100%" height="100%">
+                    <source src="{{stream_url}}" type="video/mp4">
+                    RTSP playback not supported.
+                </video>
+            </div>
+            <p class="mt-2 small-text">Frame queue length: <span id="queue-length-f">...</span></p>
+            <p class="mt-2 small-text">Inference Queue length: <span id="queue-length-i">...</span></p>
+        </div>
+
+        <!-- Negative Detections -->
+        <div class="col-md-3">
+            <h4 class="section-title text-danger">Negative Detections</h4>
+            <div class="detection-list" id="negatives">
+                <p class="text-muted small">Waiting for data...</p>
+            </div>
+        </div>
+
+        <!-- Positive Detections -->
+        <div class="col-md-3">
+            <h4 class="section-title text-success">Positive Detections</h4>
+            <div class="detection-list" id="positives">
+                <p class="text-muted small">Waiting for data...</p>
+            </div>
+        </div>
+
+    </div>
+</div>
+
+<!-- Bootstrap JS + Fetch -->
+<script src="https://cdn.jsdelivr.net/npm/bootstrap@5.3.3/dist/js/bootstrap.bundle.min.js"></script>
+<script>
+    async function fetchStatus() {
+        try {
+            const response = await fetch('/status');
+            const data = await response.json();
+
+            // Update queue length
+            document.getElementById('queue-length-f').textContent = data.stream_1_queue;
+            document.getElementById('queue-length-i').textContent = data.stream_2_queue;
+
+            // Update negatives
+            const negContainer = document.getElementById('negatives');
+            negContainer.innerHTML = '';
+            data.negatives.forEach(item => {
+                negContainer.innerHTML += `
+                    <div class="card mb-2">
+                        <img src="/gifs/${item.gif}" class="gif-preview" alt="negative">
+                        <div class="card-body p-2">
+                            <small class="text-muted">${item.timestamp}</small>
+                        </div>
+                    </div>
+                `;
+            });
+
+            // Update positives
+            const posContainer = document.getElementById('positives');
+            posContainer.innerHTML = '';
+            data.positives.forEach(item => {
+                posContainer.innerHTML += `
+                    <div class="card mb-2">
+                        <img src="/gifs/${item.gif}" class="gif-preview" alt="positive">
+                        <div class="card-body p-2">
+                            <small class="text-muted">${item.timestamp}</small>
+                        </div>
+                    </div>
+                `;
+            });
+
+        } catch (err) {
+            console.error('Error fetching status:', err);
+        }
+    }
+
+    // Fetch every 2 seconds
+    setInterval(fetchStatus, 2000);
+    fetchStatus(); // Initial call
+</script>
+</body>
+</html>

web/web.py

@@ -0,0 +1,97 @@
+import os
+import io
+import redis
+import sqlite3
+import json
+from flask import Flask, jsonify, send_from_directory, render_template, render_template_string
+
+# -----------------------------
+# ENVIRONMENT CONFIG
+# -----------------------------
+redis_host = os.getenv("redis_host", "localhost")
+redis_port = int(os.getenv("redis_port", "6379"))
+
+stream_url = os.getenv("stream_url", "/path/to/stream")
+output_folder = os.getenv("out_folder", "/app/out_folder")
+stream_label = os.getenv("stream_label", "default_stream")
+stream_label_queue = f"{stream_label}_cubes"
+
+sqlite_path = os.path.join(output_folder, f"{stream_label}_results.db")
+gif_folder = os.path.join(output_folder, stream_label)
+
+sqlite_path = os.path.join(output_folder, f"{stream_label}_results.db")
+gif_folder = os.path.join(output_folder, stream_label)
+
+# -----------------------------
+# REDIS + SQLITE CONNECTIONS
+# -----------------------------
+redis_conn = redis.Redis(host=redis_host, port=redis_port, db=0, decode_responses=False)
+
+conn = sqlite3.connect(sqlite_path, check_same_thread=False)
+cursor = conn.cursor()
+cursor.execute('''
+CREATE TABLE IF NOT EXISTS inference_results (
+    id INTEGER PRIMARY KEY AUTOINCREMENT,
+    gif_name TEXT,
+    timestamp TIMESTAMP,
+    prediction INTEGER
+)
+''')
+conn.commit()
+
+# -----------------------------
+# REDIS PUSH FUNCTION
+# -----------------------------
+def toRedis(queue_label, data):
+    print(f"Pushed data to queue: {queue_label}")
+    buffer = io.BytesIO()
+    np.savez(buffer, data=data)
+    compressed_data = buffer.getvalue()
+    return redis_conn.rpush(queue_label, compressed_data)
+
+# -----------------------------
+# FLASK SETUP
+# -----------------------------
+app = Flask(__name__,  template_folder='/app/templates')
+
+# Root returns a placeholder HTML template
+@app.route("/")
+def home():
+    return render_template("index.html", stream_url=stream_url)
+
+# JSON API endpoint for status monitoring
+@app.route("/status")
+def api_status():
+    try:
+        queue_len_1 = redis_conn.llen(stream_label)
+        queue_len_2 = redis_conn.llen(stream_label_queue)
+
+        # Fetch 10 latest positive and negative predictions
+        cursor.execute("SELECT gif_name, timestamp FROM inference_results WHERE prediction = 1 ORDER BY timestamp DESC LIMIT 10")
+        positives = [{"gif": row[0], "timestamp": row[1]} for row in cursor.fetchall()]
+
+        cursor.execute("SELECT gif_name, timestamp FROM inference_results WHERE prediction = 0 ORDER BY timestamp DESC LIMIT 10")
+        negatives = [{"gif": row[0], "timestamp": row[1]} for row in cursor.fetchall()]
+
+        return jsonify({
+            "stream_1_queue": queue_len_1,
+            "stream_2_queue": queue_len_2,
+            "positives": positives,
+            "negatives": negatives
+        })
+
+    except Exception as e:
+        return jsonify({"error": str(e)}), 500
+
+# Serve static GIFs
+@app.route("/gifs/<path:filename>")
+def serve_gif(filename):
+    return send_from_directory(gif_folder, filename)
+
+# -----------------------------
+# MAIN ENTRY
+# -----------------------------
+if __name__ == "__main__":
+    print(f"[INFO] Flask app running — monitoring queues: {stream_label}, {stream_label_queue}")
+    print(f"[INFO] Serving GIFs from: {gif_folder}")
+    app.run(host="0.0.0.0", port=8080, debug=True)