feat(bin): add cli tool

README.md

@@ -20,3 +20,8 @@ class_probabilities = predict(pose)
 gloss = predict(pose, label=True)
 ```
 
+Or in CLI, given a `.pose` file, and an ELAN file with a `SIGN` tier:
+
+```bash
+sign_language_recognition --model="kaggle_asl_signs" --pose="sign.pose" --elan="sign.eaf"
+```

pyproject.toml

@@ -10,8 +10,8 @@ dependencies = [
     "pose-format>=0.3.2",
     "numpy",
     "tflite-runtime",
-    # for inference using a huggingface hosted model
-    "huggingface-hub",
+    "pympi-ling", # Working with ELAN files in CLI
+    "huggingface-hub", # for inference using a huggingface hosted model
 ]
 
 [project.optional-dependencies]
@@ -48,4 +48,7 @@ sign_language_recognition = ["**/*.json"]
 
 [tool.pytest.ini_options]
 addopts = "-v"
-testpaths = ["sign_language_recognition"]
+testpaths = ["sign_language_recognition"]
+
+[project.scripts]
+sign_language_recognition = "sign_language_recognition.bin:main"

sign_language_recognition/bin.py

@@ -0,0 +1,55 @@
+#!/usr/bin/env python
+
+import argparse
+import importlib
+import math
+
+from pose_format.pose import Pose
+from tqdm import tqdm
+import pympi
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--model', required=True, type=str,
+                        choices=['kaggle_asl_signs'],
+                        help='model to use')
+    parser.add_argument('--pose', required=True, type=str, help='path to input pose file')
+    parser.add_argument('--elan', required=True, type=str, help='path to elan file')
+
+    return parser.parse_args()
+
+
+def main():
+    args = get_args()
+
+    module = importlib.import_module(f"sign_language_recognition.{args.model}")
+
+    print('Loading input pose...')
+    with open(args.pose, 'rb') as pose_file:
+        pose = Pose.read(pose_file.read())
+
+    print('Loading ELAN file...')
+    eaf = pympi.Elan.Eaf(file_path=args.elan, author="sign-langauge-processing/recognition")
+    sign_annotations = eaf.get_annotation_data_for_tier('SIGN')
+
+    print('Predicting signs...')
+    for segment in tqdm(sign_annotations):
+        start_frame = int((segment[0] / 1000) * pose.body.fps)
+        end_frame = math.ceil((segment[1] / 1000) * pose.body.fps)
+
+        cropped_pose = Pose(
+            header=pose.header,
+            body=pose.body[start_frame:end_frame]
+        )
+        segment[2] = module.predict(cropped_pose, label=True)
+        eaf.remove_annotation('SIGN', segment[0])
+        eaf.add_annotation('SIGN', segment[0], segment[1], segment[2])
+
+    print('Saving ELAN file...')
+    eaf.to_file(args.elan)
+
+
+if __name__ == '__main__':
+    main()
+    # python -m sign_language_recognition.bin --model="kaggle_asl_signs" --pose="sign.pose" --elan="sign.eaf"