[benchmark] evaluate the detectors on the AutoShot dataset

benchmarks/README.md

@@ -16,7 +16,14 @@ unzip BBC/videos.zip -d BBC
 rm -rf BBC/videos.zip
 ```
 
-### Evaluation
+### AutoShot
+Download `AutoShot_test.tar.gz` from [Google drive](https://drive.google.com/file/d/17diRkLlNUUjHDooXdqFUTXYje2-x4Yt6/view?usp=sharing).
+```
+tar -zxvf AutoShot.tar.gz
+rm AutoShot.tar.gz
+```
+
+## Evaluation
 To evaluate PySceneDetect on a dataset, run the following command:
 ```
 python benchmark.py -d <dataset_name> --detector <detector_name>
@@ -28,7 +35,8 @@ python evaluate.py -d BBC --detector detect-content
 
 ### Result
 The performance is computed as recall, precision, f1, and elapsed time.
-The following results indicate that ContentDetector achieves the highest performance on the BBC dataset.
+
+#### BBC
 
 |      Detector     | Recall | Precision |   F1  | Elapsed time (second) |
 |:-----------------:|:------:|:---------:|:-----:|:---------------------:|
@@ -38,6 +46,16 @@ The following results indicate that ContentDetector achieves the highest perform
 | HistogramDetector |  90.55 |   72.76   | 80.68 |         16.13         |
 | ThresholdDetector |  0.00  |   0.00    |  0.00 |         18.95         |
 
+#### AutoShot
+
+|      Detector     | Recall | Precision |   F1  | Elapsed time (second) |
+|:-----------------:|:------:|:---------:|:-----:|:---------------------:|
+|  AdaptiveDetector |  70.77 |   77.65   | 74.05 |          1.23         |
+|  ContentDetector  |  63.67 |   76.40   | 69.46 |          1.21         |
+|    HashDetector   |  56.66 |   76.35   | 65.05 |          1.16         |
+| HistogramDetector |  63.36 |   53.34   | 57.92 |          1.23         |
+| ThresholdDetector |  0.75  |   38.64   |  1.47 |          1.24         |
+
 ## Citation
 ### BBC
 ```
@@ -47,4 +65,14 @@ The following results indicate that ContentDetector achieves the highest perform
   booktitle = {Proceedings of the 23rd ACM International Conference on Multimedia},
   year      = {2015},
 }
-```
+```
+
+### AutoShot
+```
+@InProceedings{autoshot_dataset,
+  author    = {Wentao Zhu and Yufang Huang and Xiufeng Xie and Wenxian Liu and Jincan Deng and Debing Zhang and Zhangyang Wang and Ji Liu},
+  title     = {AutoShot: A Short Video Dataset and State-of-the-Art Shot Boundary Detection},
+  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
+  year      = {2023},
+}
+```

benchmarks/autoshot_dataset.py

@@ -0,0 +1,30 @@
+import glob
+import os
+
+class AutoShotDataset:
+    """
+    The AutoShot Dataset (test splits) proposed by Zhu et al. in AutoShot: A Short Video Dataset and State-of-the-Art Shot Boundary Detection
+    Link: https://openaccess.thecvf.com/content/CVPR2023W/NAS/html/Zhu_AutoShot_A_Short_Video_Dataset_and_State-of-the-Art_Shot_Boundary_Detection_CVPRW_2023_paper.html
+    The original test set consists of 200 videos, but 36 videos are missing (AutoShot/videos/<video_id>.mp4).
+    The annotated scenes are provided in corresponding files (AutoShot/annotations/<video_id>.txt)
+    """
+
+    def __init__(self, dataset_dir: str):
+        self._video_files = [
+            file for file in sorted(glob.glob(os.path.join(dataset_dir, "videos", "*.mp4")))
+        ]
+        self._scene_files = [
+            file for file in sorted(glob.glob(os.path.join(dataset_dir, "annotations", "*.txt")))
+        ]
+        for video_file, scene_file in zip(self._video_files, self._scene_files):
+            video_id = os.path.basename(video_file).split(".")[0]
+            scene_id = os.path.basename(scene_file).split(".")[0]
+            assert video_id == scene_id
+
+    def __getitem__(self, index):
+        video_file = self._video_files[index]
+        scene_file = self._scene_files[index]
+        return video_file, scene_file
+
+    def __len__(self):
+        return len(self._video_files)

benchmarks/bbc_dataset.py

@@ -20,7 +20,7 @@ def __init__(self, dataset_dir: str):
         assert len(self._video_files) == len(self._scene_files)
         for video_file, scene_file in zip(self._video_files, self._scene_files):
             video_id = os.path.basename(video_file).replace("bbc_", "").split(".")[0]
-            scene_id = os.path.basename(scene_file).split("_")[0]
+            scene_id = os.path.basename(scene_file).split("-")[0]
             assert video_id == scene_id
 
     def __getitem__(self, index):

benchmarks/benchmark.py

@@ -2,6 +2,8 @@
 import time
 
 from bbc_dataset import BBCDataset
+from autoshot_dataset import AutoShotDataset
+
 from evaluator import Evaluator
 from tqdm import tqdm
 
@@ -15,7 +17,7 @@
 )
 
 
-def make_detector(detector_name: str):
+def _make_detector(detector_name: str):
     detector_map = {
         "detect-adaptive": AdaptiveDetector(),
         "detect-content": ContentDetector(),
@@ -26,11 +28,19 @@ def make_detector(detector_name: str):
     return detector_map[detector_name]
 
 
+def _make_dataset(dataset_name: str):
+    dataset_map = {
+        "BBC": BBCDataset("BBC"),
+        "AutoShot": AutoShotDataset("AutoShot"),
+    }
+    return dataset_map[dataset_name]
+
+
 def _detect_scenes(detector_type: str, dataset):
     pred_scenes = {}
     for video_file, scene_file in tqdm(dataset):
         start = time.time()
-        detector = make_detector(detector_type)
+        detector = _make_detector(detector_type)
         pred_scene_list = detect(video_file, detector)
         elapsed = time.time() - start
         scenes = {
@@ -53,7 +63,7 @@ def _detect_scenes(detector_type: str, dataset):
 
 
 def main(args):
-    pred_scenes = _detect_scenes(detector_type=args.detector, dataset=BBCDataset("BBC"))
+    pred_scenes = _detect_scenes(detector_type=args.detector, dataset=_make_dataset(args.dataset))
     result = Evaluator().evaluate_performance(pred_scenes)
     print("Overall Results:")
     print(
@@ -65,6 +75,16 @@ def main(args):
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="Benchmarking PySceneDetect performance.")
+    parser.add_argument(
+        "--dataset",
+        type=str,
+        choices=[
+            "BBC",
+            "AutoShot",
+        ],
+        default="BBC",
+        help="Dataset name. Supported datasets are BBC and AutoShot.",
+    )
     parser.add_argument(
         "--detector",
         type=str,

benchmarks/evaluator.py

@@ -24,9 +24,8 @@ def evaluate_performance(self, pred_scenes):
             total_pred += len(pred_list)
             total_gt += len(gt_scene_list)
 
-        assert total_pred, pred_scenes
         recall = total_correct / total_gt
-        precision = total_correct / total_pred
+        precision = total_correct / total_pred if total_pred != 0 else 0
         f1 = 2 * recall * precision / (recall + precision) if (recall + precision) != 0 else 0
         avg_elapsed = mean([x["elapsed"] for x in pred_scenes.values()])
         result = {