def predict(data_path, pred_path, tempdir, proba_threshold, output_class_names,
model_profile, weight_download_region, verbose):
"""Identify species in a video.
This is a command line interface for prediction on camera trap footage. Given a path to camera trap footage,
the predict function use a deep learning model to predict the presence or absense of a variety of species of
common interest to wildlife researchers working with camera trap data.
"""
click.echo(f"Using data_path:\t{data_path}")
click.echo(f"Using pred_path:\t{pred_path}")
# Load the model into manager
manager = ModelManager(model_path=default_model_dir,
model_class='cnnensemble',
proba_threshold=proba_threshold,
tempdir=tempdir,
output_class_names=output_class_names,
model_kwargs=dict(
profile=model_profile,
download_region=weight_download_region))
# Make predictions, return a DataFrame
manager.predict(data_path, pred_path=pred_path, save=True)
def test_predict_invalid_videos(data_dir):
"""Tests whether invalid videos are correctly skipped."""
tempdir = tempfile.TemporaryDirectory()
video_directory = pathlib.Path(tempdir.name)
# create invalid (empty) videos
for i in range(2):
(video_directory / f"invalid{i:02}.mp4").touch()
# copy valid videos
test_video_path = list(data_dir.glob("*.mp4"))[0]
for i in range(2):
shutil.copy(test_video_path, video_directory / f"video{i:02}.mp4")
manager = ModelManager(
'',
model_class="cnnensemble",
output_class_names=False,
model_kwargs={"profile": "fast"},
)
predictions = manager.predict(video_directory)
assert predictions.loc[predictions.index.str.contains(
"invalid")].isnull().values.all()
assert ~predictions.loc[predictions.index.str.contains(
"video")].isnull().values.any()
tempdir.cleanup()
def test_predict_full(data_dir):
manager = ModelManager('',
model_class='cnnensemble',
output_class_names=False,
model_kwargs=dict(profile='full'))
result = manager.predict(data_dir, save=True)
result.to_csv(str(config.MODEL_DIR / 'output' / 'test_prediction.csv'))
def test_predict_fast(data_dir):
manager = ModelManager('',
model_class='cnnensemble',
output_class_names=False,
model_kwargs=dict(profile='fast'))
result = manager.predict(data_dir, save=True)
# check that duiker is most likely class (manually verified)
assert result.idxmax(axis=1).values[0] == "duiker"
result.to_csv(str(config.MODEL_DIR / 'output' / 'test_prediction.csv'))
def test_load_data(data_dir):
manager = ModelManager(
'',
model_class='cnnensemble',
output_class_names=False,
model_kwargs=dict(profile='fast'),
)
input_paths = manager.model.load_data(data_dir)
assert len(input_paths) > 0
def train(data_path, labels, site_aware, tempdir):
""" Retrain network from scratch.
Train the weights from scratch using
the provided data_path and labels.
"""
click.echo(f"Using data_path:\t{data_path}")
click.echo(f"Using labels:\t{labels}")
click.echo(f"Using tempdir:\t{tempdir}")
# Load the model into manager
manager = ModelManager(model_path=default_model_dir,
model_class='cnnensemble',
tempdir=tempdir,
model_kwargs=dict(download_weights=False,
site_aware=site_aware,
raw_video_dir=data_path,
labels_path=labels))
# Make predictions, return a DataFrame
manager.train()
def test_create_and_save(sample_model_path, sample_data_path):
manager = ModelManager(sample_model_path, model_class='sample')
# "predict" (add, multiply), return exact values since no thresh given
result = manager.predict(sample_data_path)
# 6 + 3 == 9
assert result.iloc[0].added == 9
# 6 * 3 == 18
assert result.iloc[0].multiplied == 18
# 0.3 + 0.1 == 0.4
assert result.iloc[1].added == np.float32(0.3) + np.float32(0.1)
# 0.3 * 0.1 == 0.03
assert result.iloc[1].multiplied == np.float32(0.3) * np.float32(0.1)
manager.model.save_model()
assert manager.model_path.exists()
def test_load_and_predict(sample_model_path, sample_data_path):
# load the sample model in the ModelManager
manager = ModelManager(sample_model_path,
model_class='sample',
proba_threshold=0.5)
# "predict" (add, multiply), return binary since thresh given
preds = manager.predict(sample_data_path)
# 6 + 3 == 9 >= 0.5 --> True
assert preds.iloc[0].added
# 6 * 3 == 18 >= 0.5 --> True
assert preds.iloc[0].multiplied
# 0.3 + 0.1 == 0.4 <= 0.5 --> False
assert not preds.iloc[1].added
# 0.3 * 0.1 == 0.03 <= 0.5 --> False
assert not preds.iloc[1].multiplied
def test_train():
manager = ModelManager(model_class='cnnensemble',
verbose=True,
model_kwargs=dict(download_weights=False))
manager.train(config)
def test_predict():
data_dir = config.MODEL_DIR / "input" / "raw_test"
manager = ModelManager('', model_class='cnnensemble', proba_threshold=0.5)
manager.predict(data_dir, save=True)