Skip to content

deeplabcut.pose_tracking_pytorch

Modules:

Name Description
apis
datasets
model
solver
tracking_utils
trackingutils

Functions:

Name Description
convert_coord_from_img_space_to_feature_space

If stride ==8: stride = stride * 2 elif stride == 4: stride = stride *4 elif
load_features_from_coord

Extract the deep feature at the location of the keypoint (x,y)

convert_coord_from_img_space_to_feature_space 

convert_coord_from_img_space_to_feature_space(arr, stride)

If stride ==8: stride = stride * 2 elif stride == 4: stride = stride 4 elif stride ==2: stride = stride 8.

Source code in deeplabcut/pose_tracking_pytorch/tracking_utils/preprocessing.py 
34
35
36
37
38
39
40
41
42
43
44
45
46
def convert_coord_from_img_space_to_feature_space(arr, stride):
    """If stride ==8: stride = stride * 2 elif stride == 4: stride = stride *4 elif
    stride ==2: stride = stride *8."""
    # More elegantly one can simply define:
    stride = 16

    arr = np.nan_to_num(arr).astype(np.int64)

    # take care of difference between feature map space and original image space

    arr = (arr - (stride // 2)) // stride

    return arr.astype(np.int64)

load_features_from_coord 

load_features_from_coord(feature, coords, valid_mask_for_fish=False)

Extract the deep feature at the location of the keypoint (x,y)

Source code in deeplabcut/pose_tracking_pytorch/tracking_utils/preprocessing.py 
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
def load_features_from_coord(feature, coords, valid_mask_for_fish=False):
    """Extract the deep feature at the location of the keypoint (x,y)"""
    if valid_mask_for_fish:
        mask = np.array([1, 2, 6])
        coords = coords[mask, :]

    feat_vec = np.zeros((coords.shape[0], coords.shape[1], feature.shape[-1]))

    for animal_idx in range(coords.shape[0]):
        for kpt_idx in range(coords.shape[1]):
            coord = coords[animal_idx][kpt_idx]
            x, y = coord

            vec = feature[y, x, :]
            if np.sum(coord) != 0:
                feat_vec[animal_idx][kpt_idx] = vec

    return feat_vec