Added resize before display and video record. Removed useless READMEs

README.md

@@ -27,7 +27,9 @@ The APTITUDE toolbox can be run on CPU. However, it is highly recommended to use
 <details> 
   <summary>Using Docker (recommended)</summary>
 
-If you want to infer result with the provided models or with your own model, we recommend to download the last Docker image (link available soon) or build it yourself. To do so, enter the following command. The build process should last ~15 minutes.
+If you want to infer result with the provided models or with your own model, we recommend to pull the latest Docker image available on [Dockerhub](https://hub.docker.com/repository/docker/jonathansamelson/aptitude-toolbox). 
+
+You can also build it yourself. To do so, enter the following command. The build process should last ~15 minutes.
 
 ```
 git clone https://github.com/Trusted-AI-Labs/APTITUDE_Toolbox/
@@ -118,8 +120,10 @@ The toolbox comes with two "client" scripts that can be used to produce an annot
 | -fi |--frame_interval | Interval between two detections + tracking. | 1
 | -gt |--ground_truth_path | Path to ground truth file in MOT format. It will be displayed in white. |
 | -hl |--headless | Whether the video is shown while being processed. | False
+| -ds |--display_size | Dimension of the video shown as it is processed: width,height. |
 | -rp |--record_path | Path of the output video file. |
 | -rf |--record_fps | FPS of the output video file. | 10
+| -rs |--record_size | Dimension of the recorded video: width,height. Each image is resized just before being written. |
 | -mp |--mot_path | Path to the result of tracking in MOT format.
 | -a | --async | For video files only. Whether the video reading is asynchronous. | False
 
@@ -176,7 +180,9 @@ The APTITUDE toolbox comes from the APTITUDE project (**A**pprentissage **P**rof
 
 Refer to the following repositories for more information on individual algorithms. Part of the code present in this repository was adapted from those repositories, see individual source files for details.
 
-YOLO & Tiny-YOLO inference: [OpenCV Documentation - DNN module](https://docs.opencv.org/4.5.3/d0/db7/tutorial_js_table_of_contents_dnn.html)
+YOLO & Tiny-YOLO (v2-4) inference: [OpenCV Documentation - DNN module](https://docs.opencv.org/4.5.3/d0/db7/tutorial_js_table_of_contents_dnn.html)
+
+YOLOv5 inference: [Ultralytics - YOLOv5](https://github.com/ultralytics/yolov5)
 
 SORT: [abewley - SORT](https://github.com/abewley/sort)
 

clients/main.py

@@ -12,6 +12,13 @@
     logging.getLogger("fvcore.common.checkpoint").setLevel(logging.WARNING)  # Detectron2 logger
     logging.getLogger("utils.general").setLevel(logging.WARNING)  # yolov5 logger
 
+    def dims(s):
+        try:
+            w, h = map(int, s.split(','))
+            return w, h
+        except:
+            raise argparse.ArgumentTypeError("Dimensions must be expressed as widht,height without quotes or parantheses")
+
     ap = argparse.ArgumentParser()
     ap.add_argument("-d", "--detector", required=True,
                     help="path to detector config (json file)")
@@ -26,11 +33,15 @@
     ap.add_argument("-gt", "--ground_truth_path", type=str, default=None,
                     help="path to ground truth file in MOT format.")
     ap.add_argument("-hl", "--headless", action='store_true',
-                    help="whether the video is shown as it processed")
+                    help="whether the video is shown as it is processed")
+    ap.add_argument("-ds", "--display_size", type=dims, default=None,
+                    help="dimension of the video shown as it is processed: width,height")
     ap.add_argument("-rp", "--record_path", type=str, default=None,
                     help="path of the output video file")
     ap.add_argument("-rf", "--record_fps", type=int, default=10,
                     help="fps of the output video file")
+    ap.add_argument("-rs", "--record_size", type=dims, default=None,
+                    help="dimension of the recorded video: width,height. each image is resized just before being written")
     ap.add_argument("-mp", "--mot_path", type=str, default=None,
                     help="path to the result of tracking in MOT format")
     ap.add_argument("-a", "--async", action='store_true',
@@ -49,10 +60,10 @@
 
     if os.path.isdir(args["path"]):
         tci.main(args["detector"], args["tracker"], args["classes"],
-                 args["path"], args["frame_interval"], args["record_path"], args["record_fps"],
-                 args["mot_path"], args["headless"], args["ground_truth_path"],)
+                 args["path"], args["frame_interval"], args["record_path"], args["record_fps"], args["record_size"],
+                 args["mot_path"], args["headless"], args["display_size"], args["ground_truth_path"],)
 
     else:
         tcv.main(args["detector"], args["tracker"], args["classes"],
-                 args["path"], args["frame_interval"], args["record_path"], args["record_fps"],
-                 args["mot_path"], args["headless"], args["async"], args["ground_truth_path"])
+                 args["path"], args["frame_interval"], args["record_path"], args["record_fps"], args["record_size"],
+                 args["mot_path"], args["headless"], args["display_size"], args["async"], args["ground_truth_path"])

clients/tracker_client_images.py

@@ -22,8 +22,8 @@
 thickness = 2
 
 
-def main(cfg_detect, cfg_track, cfg_classes, folder_path, frame_interval, record_path, record_fps,
-         mot_path, headless, gt_path):
+def main(cfg_detect, cfg_track, cfg_classes, folder_path, frame_interval, record_path, record_fps, record_size,
+         mot_path, headless, display_size, gt_path):
 
     log = logging.getLogger("aptitude-toolbox")
 
@@ -75,7 +75,9 @@ def main(cfg_detect, cfg_track, cfg_classes, folder_path, frame_interval, record
     frame_test = ih.get_cv2_img_from_str(os.path.join(folder_path, file_list[0]))
     H, W, _ = frame_test.shape
     if record:
-        output_video = cv2.VideoWriter(record_path, cv2.VideoWriter_fourcc(*'mp4v'), record_fps, (W, H))
+        if record_size is None:
+            record_size = (H, W)
+        output_video = cv2.VideoWriter(record_path, cv2.VideoWriter_fourcc(*'mp4v'), record_fps, record_size)
         log.debug("VideoWriter opened successfully.")
 
     if gt_path is not None and not os.path.isdir(gt_path):
@@ -185,10 +187,18 @@ def main(cfg_detect, cfg_track, cfg_classes, folder_path, frame_interval, record
         log.debug("Results bounding boxes added to the image.")
 
         if not headless:
-            cv2.imshow("Result", frame)
+            frame_display = frame
+            if display_size is not None:
+                frame_display = cv2.resize(frame_display, display_size)
+                log.debug("Frame resized for display")
+            cv2.imshow("Result", frame_display)
             log.debug("Frame displayed.")
         if record:
-            output_video.write(frame)
+            frame_record = frame
+            if record_size is not None:
+                frame_record = cv2.resize(frame_record, record_size)
+                log.debug("Frame resized for record")
+            output_video.write(frame_record)
             log.debug("Frame written to VideoWriter.")
 
         counter += 1

clients/tracker_client_video.py

@@ -20,8 +20,8 @@
 thickness = 2
 
 
-def main(cfg_detect, cfg_track, cfg_classes, video_path, frame_interval, record_path, record_fps,
-         mot_path, headless, async_flag, gt_path):
+def main(cfg_detect, cfg_track, cfg_classes, video_path, frame_interval, record_path, record_fps, record_size,
+         mot_path, headless, display_size, async_flag, gt_path):
 
     log = logging.getLogger("aptitude-toolbox")
 
@@ -84,7 +84,9 @@ def main(cfg_detect, cfg_track, cfg_classes, video_path, frame_interval, record_
     record = record_path is not None
     H, W, _ = frame.shape
     if record:
-        output_video = cv2.VideoWriter(record_path, cv2.VideoWriter_fourcc(*'mp4v'), record_fps, (W, H))
+        if record_size is None:
+            record_size = (H, W)
+        output_video = cv2.VideoWriter(record_path, cv2.VideoWriter_fourcc(*'mp4v'), record_fps, record_size)
         log.debug("VideoWriter opened successfully.")
 
     start_time = default_timer()
@@ -169,12 +171,19 @@ def main(cfg_detect, cfg_track, cfg_classes, video_path, frame_interval, record_
             log.debug("Results bounding boxes added to the image.")
 
             if not headless:
-                cv2.imshow("Result", frame)
+                frame_display = frame
+                if display_size is not None:
+                    frame_display = cv2.resize(frame_display, display_size)
+                    log.debug("Frame resized for display")
+                cv2.imshow("Result", frame_display)
                 log.debug("Frame displayed.")
             if record:
-                output_video.write(frame)
+                frame_record = frame
+                if record_size is not None:
+                    frame_record = cv2.resize(frame_record, record_size)
+                    log.debug("Frame resized for record")
+                output_video.write(frame_record)
                 log.debug("Frame written to VideoWriter.")
-
         pbar.update(1)
         counter += 1
 

pytb/utils/transformation.py

@@ -11,7 +11,7 @@
 
 def pre_process(preprocess_parameters: dict, image: np.ndarray, prev_roi: np.ndarray = None,
                 detection: Detection = None) \
-        -> Tuple[np.ndarray, Union[np.ndarray, None]]:
+        -> Tuple[np.ndarray, Union[np.ndarray, None], Union[np.ndarray, None], Union[Detection, None]]:
     if "roi" in preprocess_parameters:
         roi = prev_roi
         if prev_roi is None: