Functions — github.com/akanazawa/human

↓ 1 callersFunctionreflect_joints3d

Assumes input is 14 x 3 (the LSP skeleton subset of H3.6M)

src/util/data_utils.py:687

↓ 1 callersFunctionreflect_pose

Input is a 72-Dim vector. Global rotation (first 3) is left alone.

src/util/data_utils.py:639

↓ 1 callersFunctionrestart_until_success

(cmd)

src/evaluation/autorestart.py:6

↓ 1 callersFunctionrestore_config

Restores config settings based on saved json file.

src/evaluation/eval.py:93

↓ 1 callersFunctionrot_mat_to_axis_angle

Args: rot_matrices (24x3x3). Returns: poses_aa (72).

src/evaluation/eval_util.py:332

↓ 1 callersFunctionrotation_matrix

(args)

src/datasets/h36/read_human36m.py:94

↓ 1 callersFunctionrun_alphapose

(img_dir, out_dir)

extract_tracks.py:63

↓ 1 callersFunctionrun_poseflow

(img_dir, out_dir)

extract_tracks.py:93

↓ 1 callersFunctionsave_config

(config)

src/config.py:337

↓ 1 callersFunctionsave_neutral_shape

(labels_file, out_dir, sess)

src/datasets/threedpw/compute_neutral_shape.py:66

↓ 1 callersFunctionsave_seq_to_tfrecord_driver

(out_name, im_paths, gt2ds,

src/datasets/h36_to_tfrecords_video.py:351

↓ 1 callersFunctionsave_to_tfrecord

Saves all images to tfrecord. Args: out_name (str): Name of tfrecord. all_image_paths (list): List of list of image path nam

src/datasets/upenn_to_tfrecords_video.py:248

↓ 1 callersFunctionsave_to_tfrecord

Saves all images to tfrecord. Args: out_name (str): Name of tfrecord. im_paths (list): List of image path names. gt2

src/datasets/insta_variety_to_tfrecords.py:332

↓ 1 callersFunctionsave_to_tfrecord

Saves all images to tfrecord. Args: out_name (str): Name of tfrecord. im_paths (list): List of image path names. gt2

src/datasets/video_in_the_wild_to_tfrecords.py:333

↓ 1 callersMethodset_bgcolor

(self, color)

src/util/render/nmr_renderer.py:239

↓ 1 callersMethodset_light_dir

(self, direction, int_dir=0.8, int_amb=0.8)

src/util/render/nmr_renderer.py:234

↓ 1 callersMethodsetup_disc_summary

Sets up histograms to visualize discriminator outputs.

src/trainer_sequence_fc.py:770

↓ 1 callersMethodsetup_optimizers

(self)

src/trainer_sequence_fc.py:752

↓ 1 callersFunctionshow_seq

(seq_path, label_path)

src/datasets/upenn/read_upenn.py:127

↓ 1 callersFunctionsmooth_bbox_params

Applies median filtering and then gaussian filtering to bounding box parameters. Args: bbox_params (Nx3): [cx, cy, scale].

src/util/smooth_bbox.py:108

↓ 1 callersFunctionsplit_preds

Remove the verts from the pred dict.

src/evaluation/prediction.py:105

↓ 1 callersFunctionsubsample

Subsamples tensor, keeping frames corresponding to indices.

src/util/data_utils.py:361

↓ 1 callersFunctiontest_sequence

Tests one tube. Args: data (dict). model (Tester). pred_mode (str). has_3d (bool). Returns: Dic

src/evaluation/eval.py:196

↓ 1 callersFunctiontest_sequence_const

Runs the const vs Hal future/past test. Args: data (dict). model (Tester). has_3d (bool). Returns: erro

src/evaluation/eval.py:246

↓ 1 callersMethodtrain

(self)

src/trainer_sequence_fc.py:1022

↓ 1 callersMethodupdate_E_vars

(self)

src/datasets/resnet_extractor.py:48

↓ 1 callersMethodupdate_encoder_vars

(self)

src/evaluation/tester.py:163

↓ 1 callersMethoduse_pretrained

Returns true only if: 1. model_type is "resnet" 2. pretrained_model_path is not None 3. model_dir is NOT empty.

src/trainer_sequence_fc.py:426

↓ 1 callersMethodvisualize

Visualizes predictions as tf summary. Args: results (dict).

src/trainer_sequence_fc.py:1134

↓ 1 callersFunctionvisualize_img_orig

Visualizes the image with the ground truth keypoints and predicted keypoints in the original image space (squared). If you get out of me

src/util/render/nmr_renderer.py:333

↓ 1 callersFunctionvisualize_tfrecords

(fpaths)

src/datasets/visualize_train_tfrecords.py:20

↓ 1 callersFunctionvisualize_tfrecords

(fpaths)

src/datasets/visualize_tfrecords.py:29

↓ 1 callersFunctionvisualize_tfrecords_test

(fpaths)

src/datasets/visualize_tfrecords.py:101

Method__call__

Obtain SMPL with shape (beta) & pose (theta) inputs. Theta includes the global rotation. Args: beta: N x 10

src/tf_smpl/batch_smpl.py:89

Method__call__

Pre-processes the images by mapping the preprocess function to each frame in the sequence. Translations, scaling, and rotati

src/util/tube_augmentation.py:36

Method__call__

Args: images (TxHxWx3). image_sizes (Tx2). labels (Tx3x19). centers (Tx2). poses

src/util/tube_augmentation.py:224

Method__init__

(self, config, poses_aa, shapes, joints, kps, batch_size=None)

src/omega.py:171

Method__init__

(self, config, smpl, use_optcam=False, vis

src/omega.py:210

Method__init__

(self, weight_decay)

src/discriminators.py:10

Method__init__

(self, config)

src/data_loader_sequence.py:24

Method__init__

Args: config: carries configuration parameters (as flags), and prepare_dirs data_loader: SequenceDataLoader load- image_l

src/trainer_sequence_fc.py:55

Method__init__

pkl_path is the path to a SMPL model

src/tf_smpl/batch_smpl.py:27

Method__init__

(self, model_path, img_size=224, batch_size=64, sess=None)

src/datasets/resnet_extractor.py:14

Method__init__

(self)

src/datasets/common.py:20

Method__init__

(self, config, pretrained_resnet_path='', sequence_length=None)

src/evaluation/tester.py:26

Method__init__

(self, img_size=224, trans_max=20, delta_trans_max=3,

src/util/tube_augmentation.py:13

Method__init__

(self, img_size=224, trans_max=20, delta_trans_max=3,

src/util/tube_augmentation.py:191

Method__init__

(self, img_size=256, face_path='src/tf_smpl/smpl_faces.npy',

src/util/render/nmr_renderer.py:52

Method__len__

Returns the current sequence length. Returns: length (int).

src/omega.py:46

Functionaxis_angle_to_rot_mat

Args: poses_aa (72). Returns: rot_matrices (24x3x3).

src/evaluation/eval_util.py:318

Functionaz_fc2_groupnorm

Each block has 2 convs. So: norm --> relu --> conv --> norm --> relu --> conv --> add. Uses full convolution. Args:

src/models.py:121

Functionbatch_lrotmin

NOTE: not used bc I want to reuse R and this is simple. Output of this is used to compute joint-to-pose blend shape mapping. Equation 9 in SM

src/tf_smpl/batch_lbs.py:108

Functionbounded_random_walk

Simulates a random walk with boundary conditions. Used for data augmentation along entire tube. Based on: https://stackoverflow.com/ques

src/util/data_utils.py:787

Functioncompute_deltas_batched

Computes the change between rotation matrices using R1 * R2.T. Args: poses_prev (BxTx24x3x3). poses_curr (BxTx24x3x3).

src/ops.py:139

Functioncompute_fua_joints

(joints, new_root_joint, in_meter=False)

src/datasets/h36/read_human36m.py:250

Functioncompute_video_bbox

Given the prediction and original image info, figures out the min/max extent (bbox) of the person in the entire video. Adjust the ca

src/util/render/nmr_renderer.py:519

Functionconcat_dict_entries

Concatenates values dictionary. Args: dictionary (dict): Dict of lists with entries from appenders. appender (dict): Dict of

src/evaluation/eval_util.py:265

Functionconvert_as

(src, trg)

src/util/render/nmr_renderer.py:256

Functioncrop_and_clean_mask_to_int

(mask, x, y, w, h)

src/datasets/h36/read_human36m.py:288

Functioncrop_image

(silhs)

src/datasets/h36/read_human36m.py:264

Functionencoder_resnet

Resnet v2-50 Assumes input is [batch, height_in, width_in, channels]!! Input: - x: N x H x W x 3 - weight_decay: float - reus

src/models.py:50

Functionfc2_res

Converts pretrained (fixed) resnet features phi into movie strip. This applies 2 fc then add it to the orig as residuals. Args:

src/models.py:270

Functionget_all_files

src/util/data_utils.py:381

Methodget_all_verts

(self)

src/omega.py:325

Methodget_deltas_aa

Returns axis-aligned deltas from time t to t + 1. Args: t (int). Returns: Deltas (Bx24x3).

src/omega.py:107

Methodget_deltas_aa

(self, t=None)

src/omega.py:193

Methodget_deltas_rot

Returns deltas as rotation matrices from time t to t + 1. Args: t (int). Returns: Deltas (Bx24x3x3)

src/omega.py:119

Functionget_num_frames

(path)

src/datasets/h36/read_human36m.py:195

Functionget_params_from_omega

(smpl_model, regressor, omega, cam=None)

src/util/render/nmr_renderer.py:637

Methodget_poses_aa

Returns axis-aligned poses at time t. Args: t (int). Returns: Poses (Bx24x3).

src/omega.py:83

Methodget_shapes

(self, t=None)

src/omega.py:186

Methodget_smpl_loader_temporal

Loads real delta poses from mocap data. Returns: delta_pose_batch (BxTx216).

src/data_loader_sequence.py:221

Functionhas_num

(str)

src/datasets/h36/read_human36m.py:338

Functionjitter_center

(center, trans_max=None, rand_trans=None)

src/util/data_utils.py:512

Functionjitter_scale

(image, image_size, keypoints, center, scale_range=None, scale_factor=None)

src/util/data_utils.py:524

Methodload_pretrain

(sess)

src/trainer_sequence_fc.py:402

Functionmain

(unused_argv)

src/datasets/upenn_to_tfrecords_video.py:437

Functionmain

(unused_argv)

src/datasets/insta_variety_to_tfrecords.py:523

Functionmain

(unused_argv)

src/datasets/video_in_the_wild_to_tfrecords.py:502

Functionmain

(unused_argv)

src/datasets/smpl_to_tfrecords.py:234

Functionmain

(unused_argv)

src/datasets/h36_to_tfrecords_video.py:475

Functionmain

(unused_argv)

src/datasets/3dpw_to_tfrecords_video.py:153

Functionorthographic_proj_withz

X: B x N x 3 cam: B x 7: [sc, tx, ty, quaternions] Orth preserving the z. sc * ( x + [tx; ty]) as in HMR..

src/util/render/torch_utils.py:32

Functionorthographic_proj_withz_idrot

X: B x N x 3 cam: B x 3: [sc, tx, ty] No rotation! Orth preserving the z. sc * ( x + [tx; ty]) as in HMR..

src/util/render/torch_utils.py:11

Functionpad_image_edge

Pads image in each dimension by margin, in numpy: image_pad = np.pad(image, ((margin, margin), (ma

src/util/data_utils.py:551

Functionparse_example_proto

Parses an Example proto. Its contents are: 'image/height' : _int64_feature(height), 'image/width' : _int64_feature(w

src/util/data_utils.py:18

Functionparse_example_proto_temporal

Parses an Example proto. Its contents are: 'meta/N' : _int64_feature(N), 'image/heightwidths' : _int64_feature(

src/util/data_utils.py:119

Methodpreprocess

(data)

src/util/tube_augmentation.py:90

Functionprocess_video

(model, config, im_paths, kps, pred_dir, min_frame=0, max_frame=None)

src/evaluation/run_video.py:27

Functionproject_point_radial

(P, R, t, f, c, all_k)

src/datasets/h36/read_human36m.py:108

Functionrandom_flip

mirrors image L/R and kp, also pose if supplied

src/util/data_utils.py:582

Functionread_fua_results

(path, sbj_id, trial_id, cam_id)

src/datasets/h36/read_human36m.py:184

Functionread_silhouettes

(path, n_frames=None)

src/datasets/h36/read_human36m.py:215

Functionread_smpl_data

Parses a smpl Example proto. It's contents are: 'pose' : 72-D float 'shape' : 10-D float

src/util/data_utils.py:416

Functionread_smpl_data_temporal

Parses smpl temporal Example proto.

src/util/data_utils.py:439

Functionrescale_image

Rescales image from [0, 1] to [-1, 1] Resnet v2 style preprocessing.

src/util/data_utils.py:370

Functionrotate_img

Tensorflow's rotate does not adjust the new image size, so only work on this with square images.. image: N x N x 3 #center: 2 x 1 in

src/util/data_utils.py:702

Methodsplit_list

(list_in)

src/data_loader_sequence.py:82

Functionvisualize_mesh_og

Visualize mesh in original image space. If you get out of memory error, make max_img_size smaller. If crop_cam and bbox is not None,

src/util/render/nmr_renderer.py:422

Functions299 in github.com/akanazawa/human_dynamics