tensor input dump and UI to classify

README.md

@@ -238,6 +238,15 @@ objects:
       # Optional: minimum decimal percentage for tracked object's computed score to be considered a true positive (default: shown below)
       threshold: 0.85
 
+# Configuraiton for saving data for training.
+# Set to TRUE to save inputs to tensorflow. Note that every input is saved, so this will generate a lot of data.
+saveTensorInputs: false
+# Path to save tensor inputs to.
+saveTensorPath: /clips/tensorInputs
+# Path the categorizer UI will save categorized inputs to.
+# The categorizer UI can be accessed at http://<frigate url>/tensorInputUi
+saveTensorCategorizedPath: /clips/tensorCategorized
+
 # Required: configuration section for cameras
 cameras:
   # Required: name of the camera

detect_objects.py

@@ -15,8 +15,9 @@ import multiprocessing as mp
 import subprocess as sp
 import numpy as np
 import logging
-from flask import Flask, Response, make_response, jsonify, request
+from flask import Flask, Response, make_response, jsonify, request, send_from_directory
 import paho.mqtt.client as mqtt
+import shutil
 
 from frigate.video import capture_camera, track_camera, get_ffmpeg_input, get_frame_shape, CameraCapture, start_or_restart_ffmpeg
 from frigate.object_processing import TrackedObjectProcessor
@@ -121,8 +122,9 @@ class FrigateWatchdog(threading.Thread):
                     camera_process['process_fps'].value = 0.0
                     camera_process['detection_fps'].value = 0.0
                     camera_process['read_start'].value = 0.0
-                    process = mp.Process(target=track_camera, args=(name, self.config,
-                        self.detection_queue, self.out_events[name], self.tracked_objects_queue, camera_process, self.stop_event))
+                    process = mp.Process(target=track_camera, args=(name, self.config[name],
+                        self.detection_queue, self.out_events[name], self.tracked_objects_queue, camera_process, 
+                        self.stop_event, CONFIG))
                     process.daemon = True
                     camera_process['process'] = process
                     process.start()
@@ -265,7 +267,7 @@ def main():
         camera_process_info[name]['capture_process'] = capture_process
 
         camera_process = mp.Process(target=track_camera, args=(name, config,
-            detection_queue, out_events[name], tracked_objects_queue, camera_process_info[name], stop_event))
+            detection_queue, out_events[name], tracked_objects_queue, camera_process_info[name], stop_event, CONFIG))
         camera_process.daemon = True
         camera_process_info[name]['process'] = camera_process
 
@@ -417,7 +419,7 @@ def main():
             return response
         else:
             return "Camera named {} not found".format(camera_name), 404
-            
+        
     def imagestream(camera_name, fps, height):
         while True:
             # max out at specified FPS
@@ -433,6 +435,110 @@ def main():
             yield (b'--frame\r\n'
                 b'Content-Type: image/jpeg\r\n\r\n' + jpg.tobytes() + b'\r\n\r\n')
 
+    @app.route('/tensorInput/<path:path>')
+    def send_tensor_image(path):
+        return send_from_directory(CONFIG["saveTensorPath"], path)
+
+    @app.route('/tensorInputClassify/<path:path>/<category>')
+    def classify_tensor_image(path, category):
+        if category == "delete":
+            os.unlink(os.path.join(CONFIG["saveTensorPath"], path))
+            return "deleted"
+        else:
+            category_path = os.path.join(CONFIG["saveTensorCategorizedPath"], category)
+            os.makedirs(category_path, exist_ok=True)
+            shutil.move(os.path.join(CONFIG["saveTensorPath"], path), os.path.join(category_path, path))
+            return "ok"
+
+    @app.route('/tensorInputNext')
+    def get_next_tensor_image():
+        file = None
+        for root, dirs, files in os.walk(CONFIG["saveTensorPath"]):
+            for name in files:
+                file = name
+                break
+        return "No more inputs" if file is None else file
+
+    @app.route('/tensorInputUi')
+    def get_tensor_ui():
+        return """<html>
+<head>
+<title>Simple classification UI</title>
+<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/water.css@2/out/water.css">
+</head>
+
+<script type="text/javascript">
+var current = ""
+let categories = {}
+
+const getNext = async () => {
+	let response = await fetch('./tensorInputNext')
+  current = await response.text()
+  if (current === "No more inputs") {
+      // probably make this nicer in the future
+      alert("All done, no more inputs");
+  }
+  document.getElementById("image").src = `./tensorInput/${current}`
+}
+
+const addCategory = (name) => {
+	let row = document.getElementById("categories").insertRow(-1);
+  let i = -1;
+	let letter = ""
+  while (letter in categories || i == -1) {
+  	i++;
+    if (i < name.length) {
+    	letter = name.toLowerCase().charAt(i);
+    } else {
+    	//out of letters, pick some random key until it works
+       const alphabet = "abcdefghijklmnopqrstuvwxyz1234567890,./;'[]\-=`'"
+       letter = alphabet[Math.floor(Math.random() * alphabet.length)]
+    }
+  }
+  
+  categories[letter] = name
+  row.insertCell().appendChild(document.createTextNode(letter));
+  row.insertCell().appendChild(document.createTextNode(name));
+}
+
+const classify = async (key) => {
+	await fetch(`./tensorInputClassify/${current}/${categories[key].toLowerCase()}`)
+  await getNext()
+}
+
+window.onload=()=> {
+	getNext();
+  // add default delete category
+  addCategory("Delete");
+  document.getElementById("add_category").onclick = () => {
+    addCategory(document.getElementById("category").value)
+    document.getElementById("category").value = ""
+    return false;
+  }
+}
+
+document.addEventListener('keypress', event => {
+		if (event.key in categories && document.getElementById("category") != document.activeElement) {
+    	classify(event.key);
+    }
+  });
+</script>
+
+<img style="height:300px;width:300px;" id="image"/>
+<hr>
+<form>
+ <label for="category">New Category</label>
+ <input type="text" id="category"> 
+ <button id="add_category">Add Category</button>
+</form>
+<table id="categories"> 
+<thead> 
+<tr> <th>Key</th> <th>Category</th></thead><tbody> 
+</tbody> </table>
+
+</html>
+        """
+
     app.run(host='0.0.0.0', port=WEB_PORT, debug=False)
 
     object_processor.join()

frigate/edgetpu.py

@@ -10,6 +10,7 @@ import numpy as np
 import tflite_runtime.interpreter as tflite
 from tflite_runtime.interpreter import load_delegate
 from frigate.util import EventsPerSecond, listen, SharedMemoryFrameManager
+import cv2
 
 def load_labels(path, encoding='utf-8'):
   """Loads labels from file (with or without index numbers).
@@ -159,7 +160,7 @@ class EdgeTPUProcess():
         self.detect_process.start()
 
 class RemoteObjectDetector():
-    def __init__(self, name, labels, detection_queue, event):
+    def __init__(self, name, labels, detection_queue, event, config):
         self.labels = load_labels(labels)
         self.name = name
         self.fps = EventsPerSecond()
@@ -169,6 +170,7 @@ class RemoteObjectDetector():
         self.np_shm = np.ndarray((1,300,300,3), dtype=np.uint8, buffer=self.shm.buf)
         self.out_shm = mp.shared_memory.SharedMemory(name=f"out-{self.name}", create=False)
         self.out_np_shm = np.ndarray((20,6), dtype=np.float32, buffer=self.out_shm.buf)
+        self.config = config
     
     def detect(self, tensor_input, threshold=.4):
         detections = []
@@ -177,6 +179,14 @@ class RemoteObjectDetector():
         self.np_shm[:] = tensor_input[:]
         self.event.clear()
         self.detection_queue.put(self.name)
+
+        if self.config["saveTensorInputs"]:
+            root_path = self.config['saveTensorPath']
+            os.makedirs(root_path, exist_ok=True)
+
+            file_path = os.path.join(root_path, f"{self.name}.{datetime.datetime.now().timestamp()}.jpg")
+            cv2.imwrite(file_path, tensor_input[0])
+
         result = self.event.wait(timeout=10.0)
 
         # if it timed out

frigate/video.py

@@ -241,7 +241,7 @@ def capture_camera(name, config, process_info, stop_event):
     camera_watchdog.start()
     camera_watchdog.join()
 
-def track_camera(name, config, detection_queue, result_connection, detected_objects_queue, process_info, stop_event):
+def track_camera(name, config, detection_queue, result_connection, detected_objects_queue, process_info, stop_event, global_config):
     listen()
 
     frame_queue = process_info['frame_queue']
@@ -275,7 +275,7 @@ def track_camera(name, config, detection_queue, result_connection, detected_obje
         mask[:] = 255
 
     motion_detector = MotionDetector(frame_shape, mask, resize_factor=6)
-    object_detector = RemoteObjectDetector(name, '/labelmap.txt', detection_queue, result_connection)
+    object_detector = RemoteObjectDetector(name, '/labelmap.txt', detection_queue, result_connection, global_config)
 
     object_tracker = ObjectTracker(10)