Update contents

.pre-commit-config.yaml

@@ -22,4 +22,4 @@ repos:
     rev: v4.0.1
     hooks:
       - id: check-added-large-files
-        args: ["--maxkb=100"]
+        args: ["--maxkb=500"]

LICENSE

@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2020 Ozora Ogino (ozora-ogino)
+Copyright (c) 2021 Ozora Ogino (ozora-ogino)
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

README.md

@@ -1,10 +1,57 @@
+<h1 align="center">
+  TensorFlow Light Human Tracking
+</h1>
 
-## SORT
+![](./outputs/trim10s.mp4_yolov5l-fp16.tflite.mp4)
 
-> A simple online and realtime tracking algorithm for 2D multiple object tracking in video sequences.
+The motivation of TensorFlow Lite Human Tracking is developing person tacking system for edge camera.
+For example, count the number of visitors in somewhere.
 
-https://github.com/abewley/sort
+To track and detect people over frames, DeepSORT is adopted.
 
-## Dataset
+For the detail about DeepSORT, refer [this great article](https://medium.com/augmented-startups/deepsort-deep-learning-applied-to-object-tracking-924f59f99104).
 
-https://www.kaggle.com/ashayajbani/oxford-town-centre/version/4?select=TownCentreXVID.mp4
+
+Currently [YOLOv5](https://github.com/ultralytics/yolov5) models are supported for object detection model.
+To get YOLOv5 tflite model, see [`models/README.md`](./models/README.md)
+
+## <div align="center">Quick Start Example</div>
+
+```bash
+git clone [email protected]:ozora-ogino/tflite-human-tracking.git
+cd tflite-human-tracking
+python main.py --src ./data/<YOUR_VIDEO_FILE>.mp4 --model ./models/<YOLOV5_MODEL>.tflite
+```
+
+### Docker
+
+```bash
+./build_image.sh
+./run.sh ./data/<YOUR_VIDEO_FILE>.mp4 ./models/<YOLOV5_MODEL>.tflite
+```
+
+Then you can see the results in `outputs` folder.
+
+
+### Dataset
+The example video on top of here is [TownCentreXVID](https://www.kaggle.com/ashayajbani/oxford-town-centre/version/4?select=TownCentreXVID.mp4).
+You can download it from the link (kaggle).
+
+I recoomend to trim it for about 10s because it's too big for testing.
+
+
+## <div align="center">Citations</div>
+
+### SORT
+
+```
+@inproceedings{Bewley2016_sort,
+  author={Bewley, Alex and Ge, Zongyuan and Ott, Lionel and Ramos, Fabio and Upcroft, Ben},
+  booktitle={2016 IEEE International Conference on Image Processing (ICIP)},
+  title={Simple online and realtime tracking},
+  year={2016},
+  pages={3464-3468},
+  keywords={Benchmark testing;Complexity theory;Detectors;Kalman filters;Target tracking;Visualization;Computer Vision;Data Association;Detection;Multiple Object Tracking},
+  doi={10.1109/ICIP.2016.7533003}
+}
+```

models/README.md

@@ -0,0 +1,15 @@
+## YOLOv5 TF Light
+
+To get yolov5 tflite models, you can use [`yolov5/export.py`](https://github.com/ultralytics/yolov5/blob/master/export.py).
+
+For example;
+
+```bash
+git clone [email protected]:ultralytics/yolov5.git
+cd yolov5
+pip install -r requirements.txt
+
+python export.py --weights yolov5n.pt --include  tflite
+```
+
+For mode details, see the [official release note](https://github.com/ultralytics/yolov5/releases).

outputs/.gitignore

@@ -0,0 +1,4 @@
+**
+!.gitignore
+!trim10s.mp4_yolov5l-fp16.tflite.jpg
+!trim10s.mp4_yolov5l-fp16.tflite.mp4

src/main.py

@@ -98,8 +98,11 @@ def main(
         # Executed only first time.
         if writer is None:
             # Initialize video writer.
+            model_name = os.path.basename(model).split(".")[0]
+            video_name = os.path.basename(src).split(".")[0]
             codecs = {"mp4": "MP4V", "avi": "MJPG"}
-            output_video = os.path.join(dest, f"result.{video_fmt}")
+            basename = f"{video_name}_{model_name}"
+            output_video = os.path.join(dest, f"{basename}.{video_fmt}")
             fourcc = cv2.VideoWriter_fourcc(*codecs[video_fmt])
             writer = cv2.VideoWriter(output_video, fourcc, 30, (frame.shape[1], frame.shape[0]), True)
 
@@ -109,7 +112,7 @@ def main(
             print(f"Estimated total time: {second_per_frame * total_frames:.4f}")
 
         # Save frame as an image and video.
-        cv2.imwrite(os.path.join(dest, "detect.jpg"), frame)
+        cv2.imwrite(os.path.join(dest, f"{basename}.jpg"), frame)
         writer.write(frame)
 
     writer.release()
@@ -123,7 +126,7 @@ def main(
     parser.add_argument("--src", help="Path to video source.", default="./data/TownCentreXVID.mp4")
     parser.add_argument("--dest", help="Path to output directory", default="./outputs/")
     parser.add_argument("--model", help="Path to YOLOv5 tflite file", default="./models/yolov5n6-fp16.tflite")
-    parser.add_argument("--video-fmt", help="Format of output video file.", choices=["mp4", "avi"], default="avi")
+    parser.add_argument("--video-fmt", help="Format of output video file.", choices=["mp4", "avi"], default="mp4")
     parser.add_argument("--confidence", type=float, default=0.2, help="Confidence threshold.")
     parser.add_argument("--iou-threshold", type=float, default=0.2, help="IoU threshold for NMS.")
     args = vars(parser.parse_args())