提交 584ae970 编写于 作者: E Evgeny Izutov

Updated README

上级 7012d884
...@@ -143,11 +143,18 @@ PYTHONPATH=$PYTHONPATH:$CAFFE_ROOT/python $CAFFE_ROOT/build/tools/caffe train -- ...@@ -143,11 +143,18 @@ PYTHONPATH=$PYTHONPATH:$CAFFE_ROOT/python $CAFFE_ROOT/build/tools/caffe train --
``` ```
### Action Recognition model evaluation ### Action Recognition model evaluation
To evaluate the quality of trained Action Recognition model on your test data you can use provided script. To do this you need the file with testing tasks in the same format as for training stage (see [this](#train-tasks-file-format) section). To evaluate the quality of trained Action Recognition model on your test data you can use provided script. To do this you need the file with testing tasks in the same format as for training stage (see [this](#train-tasks-file-format) section). The model can be evaluated in two modes:
1. Frame independent evaluation:
``` ```
PYTHONPATH=$PYTHONPATH:$CAFFE_ROOT/python python2 $CAFFE_ROOT/python/action_metrics.py -t path_to_testing_tasks.txt -p $CAFFE_ROOT/models/person_detection_action_recognition/stage2__action_regognition/deploy.prototxt -w action_detection_best_snapshot_path.caffemodel PYTHONPATH=$PYTHONPATH:$CAFFE_ROOT/python python2 $CAFFE_ROOT/python/action_metrics.py -t path_to_testing_tasks.txt -p $CAFFE_ROOT/models/person_detection_action_recognition/stage2__action_regognition/deploy.prototxt -w action_detection_best_snapshot_path.caffemodel
``` ```
2. Event-based evaluation:
```
PYTHONPATH=$PYTHONPATH:$CAFFE_ROOT/python python2 $CAFFE_ROOT/python/action_event_metrics.py -t path_to_testing_tasks.txt -p $CAFFE_ROOT/models/person_detection_action_recognition/stage2__action_regognition/deploy.prototxt -w action_detection_best_snapshot_path.caffemodel
```
### Conversion to MO-compatible format ### Conversion to MO-compatible format
1. Run script to convert weights into MO-compatible format: 1. Run script to convert weights into MO-compatible format:
......
...@@ -692,9 +692,9 @@ def main(): ...@@ -692,9 +692,9 @@ def main():
parser.add_argument('--compute_mode', type=str, choices=['CPU', 'GPU'], default='GPU', parser.add_argument('--compute_mode', type=str, choices=['CPU', 'GPU'], default='GPU',
help='Caffe compute mode: CPU or GPU') help='Caffe compute mode: CPU or GPU')
parser.add_argument('--gpu_id', '-g', type=int, default=0, choices=range(8), help='GPU id') parser.add_argument('--gpu_id', '-g', type=int, default=0, choices=range(8), help='GPU id')
parser.add_argument('--min_detection_conf', '-dc', type=float, default=0.1, help='Min detection conf') parser.add_argument('--min_detection_conf', '-dc', type=float, default=0.4, help='Min detection conf')
parser.add_argument('--min_action_conf', '-ac', type=float, default=0.7, help='Min action conf') parser.add_argument('--min_action_conf', '-ac', type=float, default=0.7, help='Min action conf')
parser.add_argument('--min_action_length', type=float, default=0.0, help='Min action duration (s)') parser.add_argument('--min_action_length', type=float, default=1.0, help='Min action duration (s)')
parser.add_argument('--window_size', type=float, default=1.0, help='Smooth window size (s)') parser.add_argument('--window_size', type=float, default=1.0, help='Smooth window size (s)')
parser.add_argument('--out_dir', type=str, default='', help='Path to save smoothed annotation') parser.add_argument('--out_dir', type=str, default='', help='Path to save smoothed annotation')
args = parser.parse_args() args = parser.parse_args()
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册