enhance demo program; autocrlf

MVIG-SJTU · Sep 24, 2020 · 8b43521 · 8b43521
1 parent bcb6b16
commit 8b43521
Show file tree

Hide file tree

Showing 8 changed files with 1,683 additions and 1,592 deletions.
diff --git a/demo/README.md b/demo/README.md
@@ -8,7 +8,7 @@ To run this demo, make sure that you install all requirements following [INSTALL
 
 1. Download the object detection model manually: **yolov3-spp.weights** ([Google Drive](https://drive.google.com/open?id=1T13mXnPLu8JRelwh60BRR21f2TlGWBAM)). Place it into `data/models/detector_models`.
 2. Download the person tracking model manually: **jde.uncertainty.pt** ([Google Drive](https://drive.google.com/open?id=1IJSp_t5SRlQarFClrRolQzSJ4K5xZIqm)). Place it into `data/models/detector_models`.
-3. Please download our action models. Place them into ```data/models/aia_models```. All models are available in the [Model Zoo](../README.md#model-zoo).
+3. Please download our action models. Place them into ```data/models/aia_models```. All models are available in [MODEL_ZOO.md](../MODEL_ZOO.md).
 4. We also provide a practical model ([Google Drive](https://drive.google.com/open?id=1CudK8w0d2_5r73_tnyAY1Fnwd78hce3M)) trained on 15 common action categories in AVA. This 
 model achieves about 70 mAP on these categories. Please use [`resnet101_8x8f_denseserial.yaml`](../config_files/resnet101_8x8f_denseserial.yaml)
 and eable `--common-cate` to apply this model.

diff --git a/demo/action_predictor.py b/demo/action_predictor.py
diff --git a/demo/demo.py b/demo/demo.py
@@ -1,185 +1,194 @@
-# -----------------------------------------------------
-# Copyright (c) Shanghai Jiao Tong University. All rights reserved.
-# Written by Xinzhi MU ([email protected])
-# -----------------------------------------------------
-
-import argparse
-from time import sleep
-from itertools import count
-from tqdm import tqdm
-
-import torch
-
-from visualizer import AVAVisualizer
-from action_predictor import AVAPredictorWorker
-
-#pytorch issuse #973
-import resource
-
-rlimit = resource.getrlimit(resource.RLIMIT_NOFILE)
-resource.setrlimit(resource.RLIMIT_NOFILE, (rlimit[1], rlimit[1]))
-
-def main():
-    parser = argparse.ArgumentParser(description='Action Detection Demo')
-    parser.add_argument(
-        "--webcam",
-        dest="webcam",
-        help="Use webcam as input",
-        action="store_true",
-    )
-    parser.add_argument(
-        "--video-path",
-        default="input.mp4",
-        help="The path to the input video",
-        type=str,
-    )
-    parser.add_argument(
-        "--output-path",
-        default="output.mp4",
-        help="The path to the video output",
-        type=str,
-    )
-    parser.add_argument(
-        "--cpu",
-        dest="cpu",
-        help="Use cpu",
-        action="store_true",
-    )
-    parser.add_argument(
-        "--cfg-path",
-        default="../config_files/resnet101_8x8f_denseserial.yaml",
-        help="The path to the cfg file",
-        type=str,
-    )
-    parser.add_argument(
-        "--weight-path",
-        default="../data/models/aia_models/resnet101_8x8f_denseserial.pth",
-        help="The path to the model weights",
-        type=str,
-    )
-    parser.add_argument(
-        "--visual-threshold",
-        default=0.5,
-        help="The threshold of visualizer",
-        type=float,
-    )
-    parser.add_argument(
-        "--start",
-        default=0,
-        help="Start reading video at which millisecond",
-        type=int,
-    )
-    parser.add_argument(
-        "--duration",
-        default=-1,
-        help="The duration of detection",
-        type=int,
-    )
-    parser.add_argument(
-        "--detect-rate",
-        default=4,
-        help="Rate(fps) to update action labels",
-        type=int
-    )
-    parser.add_argument(
-        "--common-cate",
-        default=False,
-        help="Using common category model",
-        action="store_true"
-    )
-    parser.add_argument(
-        "--hide-time",
-        default=False,
-        help="Not show the timestamp at the corner",
-        action="store_true"
-    )
-    parser.add_argument(
-        "--tracker-box-thres",
-        default=0.1,
-        help="The box threshold for tracker",
-        type=float,
-    )
-    parser.add_argument(
-        "--tracker-nms-thres",
-        default=0.4,
-        help="The nms threshold for tracker",
-        type=float,
-    )
-
-    args = parser.parse_args()
-
-    args.input_path = 0 if args.webcam else args.video_path
-    args.device = torch.device("cpu" if args.cpu else "cuda")
-    args.realtime = True if args.webcam else False
-
-    # Configuration for Tracker. Currently Multi-gpu is not supported
-    args.gpus = "0"
-    args.gpus = [int(i) for i in args.gpus.split(',')] if torch.cuda.device_count() >= 1 else [-1]
-    args.min_box_area = 0
-    args.tracking = True
-    args.detector = "tracker"
-    args.debug = False
-
-    if args.webcam:
-        print('Starting webcam demo, press Ctrl + C to terminate...')
-    else:
-        print('Starting video demo, video path: {}'.format(args.video_path))
-
-    # Initialise Visualizer
-    video_writer = AVAVisualizer(
-        args.input_path,
-        args.output_path,
-        args.realtime,
-        args.start,
-        args.duration,
-        (not args.hide_time),
-        confidence_threshold = args.visual_threshold,
-        common_cate = args.common_cate,
-    )
-
-    torch.multiprocessing.set_start_method('forkserver', force=True)
-    torch.multiprocessing.set_sharing_strategy('file_system')
-
-    ava_predictor_worker = AVAPredictorWorker(args)
-
-    try:
-        for i in tqdm(count()):
-            with torch.no_grad():
-                (orig_img, boxes, scores, ids) = ava_predictor_worker.read_track()
-
-                if orig_img is None:
-                    if not args.realtime:
-                        ava_predictor_worker.compute_prediction()
-                    break
-
-                if args.realtime:
-                    result = ava_predictor_worker.read()
-                    flag = video_writer.realtime_write_frame(result, orig_img, boxes, scores, ids)
-                    if not flag:
-                        break
-                else:
-                    video_writer.send_track((boxes, ids))
-    except KeyboardInterrupt:
-        print("Keyboard Interrupted")
-
-    if not args.realtime:
-        video_writer.send_track("DONE")
-        while True:
-            result = ava_predictor_worker.read()
-            if result is None:
-                sleep(0.1)
-                continue
-            if result == "done":
-                break
-
-            video_writer.send(result)
-
-        video_writer.send("DONE")
-        print("Wait for writer process to finish...")
-        video_writer.progress_bar(i)
-
-    video_writer.close()
-    ava_predictor_worker.terminate()
-
-if __name__ == "__main__":
-    main()
+# -----------------------------------------------------
+# Copyright (c) Shanghai Jiao Tong University. All rights reserved.
+# Written by Xinzhi MU ([email protected])
+# -----------------------------------------------------
+
+import argparse
+from time import sleep
+from itertools import count
+from tqdm import tqdm
+
+import torch
+
+from visualizer import AVAVisualizer
+from action_predictor import AVAPredictorWorker
+
+#pytorch issuse #973
+import resource
+
+rlimit = resource.getrlimit(resource.RLIMIT_NOFILE)
+resource.setrlimit(resource.RLIMIT_NOFILE, (rlimit[1], rlimit[1]))
+
+def main():
+    parser = argparse.ArgumentParser(description='Action Detection Demo')
+    parser.add_argument(
+        "--webcam",
+        dest="webcam",
+        help="Use webcam as input",
+        action="store_true",
+    )
+    parser.add_argument(
+        "--video-path",
+        default="input.mp4",
+        help="The path to the input video",
+        type=str,
+    )
+    parser.add_argument(
+        "--output-path",
+        default="output.mp4",
+        help="The path to the video output",
+        type=str,
+    )
+    parser.add_argument(
+        "--cpu",
+        dest="cpu",
+        help="Use cpu",
+        action="store_true",
+    )
+    parser.add_argument(
+        "--cfg-path",
+        default="../config_files/resnet101_8x8f_denseserial.yaml",
+        help="The path to the cfg file",
+        type=str,
+    )
+    parser.add_argument(
+        "--weight-path",
+        default="../data/models/aia_models/resnet101_8x8f_denseserial.pth",
+        help="The path to the model weights",
+        type=str,
+    )
+    parser.add_argument(
+        "--visual-threshold",
+        default=0.5,
+        help="The threshold of visualizer",
+        type=float,
+    )
+    parser.add_argument(
+        "--start",
+        default=0,
+        help="Start reading video at which millisecond",
+        type=int,
+    )
+    parser.add_argument(
+        "--duration",
+        default=-1,
+        help="The duration of detection",
+        type=int,
+    )
+    parser.add_argument(
+        "--detect-rate",
+        default=4,
+        help="Rate(fps) to update action labels",
+        type=int
+    )
+    parser.add_argument(
+        "--common-cate",
+        default=False,
+        help="Using common category model",
+        action="store_true"
+    )
+    parser.add_argument(
+        "--hide-time",
+        default=False,
+        help="Not show the timestamp at the corner",
+        action="store_true"
+    )
+    parser.add_argument(
+        "--tracker-box-thres",
+        default=0.1,
+        help="The box threshold for tracker",
+        type=float,
+    )
+    parser.add_argument(
+        "--tracker-nms-thres",
+        default=0.4,
+        help="The nms threshold for tracker",
+        type=float,
+    )
+
+    args = parser.parse_args()
+
+    args.input_path = 0 if args.webcam else args.video_path
+    args.device = torch.device("cpu" if args.cpu else "cuda")
+    args.realtime = True if args.webcam else False
+
+    # Configuration for Tracker. Currently Multi-gpu is not supported
+    args.gpus = "0"
+    args.gpus = [int(i) for i in args.gpus.split(',')] if torch.cuda.device_count() >= 1 else [-1]
+    args.min_box_area = 0
+    args.tracking = True
+    args.detector = "tracker"
+    args.debug = False
+
+    if args.webcam:
+        print('Starting webcam demo, press Ctrl + C to terminate...')
+    else:
+        print('Starting video demo, video path: {}'.format(args.video_path))
+
+    # Initialise Visualizer
+    video_writer = AVAVisualizer(
+        args.input_path,
+        args.output_path,
+        args.realtime,
+        args.start,
+        args.duration,
+        (not args.hide_time),
+        confidence_threshold = args.visual_threshold,
+        common_cate = args.common_cate,
+    )
+
+    torch.multiprocessing.set_start_method('forkserver', force=True)
+    torch.multiprocessing.set_sharing_strategy('file_system')
+
+    ava_predictor_worker = AVAPredictorWorker(args)
+    pred_done_flag = False
+
+    print("Showing tracking progress bar (in fps). Other processes are running in the background.")
+    try:
+        for i in tqdm(count(), desc="Tracker Progress", unit=" frame"):
+            with torch.no_grad():
+                (orig_img, boxes, scores, ids) = ava_predictor_worker.read_track()
+
+                if orig_img is None:
+                    if not args.realtime:
+                        ava_predictor_worker.compute_prediction()
+                    break
+
+                if args.realtime:
+                    result = ava_predictor_worker.read()
+                    flag = video_writer.realtime_write_frame(result, orig_img, boxes, scores, ids)
+                    if not flag:
+                        break
+                else:
+                    video_writer.send_track((boxes, ids))
+                    while not pred_done_flag:
+                        result = ava_predictor_worker.read()
+                        if result is None:
+                            break
+                        elif result == "done":
+                            pred_done_flag = True
+                        else:
+                            video_writer.send(result)
+    except KeyboardInterrupt:
+        print("Keyboard Interrupted")
+
+    if not args.realtime:
+        video_writer.send_track("DONE")
+        while not pred_done_flag:
+            result = ava_predictor_worker.read()
+            if result is None:
+                sleep(0.1)
+            elif result == "done":
+                pred_done_flag = True
+            else:
+                video_writer.send(result)
+
+        video_writer.send("DONE")
+        tqdm.write("Showing video writer progress (in fps).")
+        video_writer.progress_bar(i)
+
+    video_writer.close()
+    ava_predictor_worker.terminate()
+
+if __name__ == "__main__":
+    main()