Function get_aug_config

util/preprocessing.py:112–171

(data_name)

def get_aug_config(data_name):
    scale_factor = 0.25
    rot_factor = 30
    color_factor = 0.2
    crop_factor = 0.1
    
    if data_name == 'GTA_Human2':
        sample_ratio = 0.5
        sample_prob = 0.5
    elif data_name == 'AGORA_MM':
        sample_ratio = 0.5
        sample_prob = 0.7
    elif data_name == 'BEDLAM':
        sample_ratio = 0.6
        sample_prob = 0.7 
    elif data_name == 'COCO_NA':
        sample_ratio = 0.6
        sample_prob = 0.7
    elif data_name == 'CrowdPose':
        sample_ratio = 0.5
        sample_prob = 0.5
    elif data_name == 'PoseTrack':
        sample_ratio = 0.5
        sample_prob = 0.3
    elif data_name == 'UBody_MM':
        sample_ratio = 0.5
        sample_prob = 0.3        
    elif data_name == 'ARCTIC':
        sample_ratio = 0.5
        sample_prob = 0.3
    elif data_name == 'RICH':
        sample_ratio = 0.5
        sample_prob = 0.3
    elif data_name == 'EgoBody_Egocentric':
        sample_ratio = 0.5
        sample_prob = 0.3
    elif data_name == 'EgoBody_Kinect':
        sample_ratio = 0.5
        sample_prob = 0.3
    else:
        sample_ratio = 0.5
        sample_prob = 0.3
    scale = np.clip(np.random.randn(), -1.0, 1.0) * scale_factor + 1.0
    rot = np.clip(np.random.randn(), -2.0,
                  2.0) * rot_factor if random.random() <= 0.6 else 0
    c_up = 1.0 + color_factor
    c_low = 1.0 - color_factor
    color_scale = np.array([
        random.uniform(c_low, c_up),
        random.uniform(c_low, c_up),
        random.uniform(c_low, c_up)
    ])
    do_flip = random.random() < 0.5
    crop_hw = np.array([
        0.2 - np.random.rand() * crop_factor, 0.2 - np.random.rand() * crop_factor
    ])
    # crop_hw = np.array([
    #     0.3 - np.random.rand() * crop_factor, 0.3 - np.random.rand() * crop_factor