Functions — github.com/akanazawa/human

↓ 1 callersFunctionadd_to_tfrecord

Adds all information from a subject-sequence-camera tuple to a tfrecord.

src/datasets/upenn_to_tfrecords_video.py:162

↓ 1 callersFunctionadd_to_tfrecord

Adds all information from a subject-sequence-camera tuple to a tfrecord.

src/datasets/insta_variety_to_tfrecords.py:191

↓ 1 callersFunctionadd_to_tfrecord

Adds all information from a subject-sequence-camera tuple to a tfrecord.

src/datasets/video_in_the_wild_to_tfrecords.py:192

↓ 1 callersFunctionadd_to_tfrecord

gt2d is T x 14 x 2 --> We need T for each for cropping gt3d is T x 14 x 3 cam is T x 3 [f, px, py] --> We need T for each for cropping

src/datasets/h36_to_tfrecords_video.py:192

↓ 1 callersFunctionadd_to_tfrecord

Args: image_paths (N). gt2ds (Nx19x3). gt3ds (Nx14x3). poses (Nx72). shape (10). coder (ImageCod

src/datasets/make_test_tfrecords.py:84

↓ 1 callersFunctionaz_fc_block2

Full convolutions not separable!! same as az_fc_block, but residual connections is proper Kaiming style of BN -> Relu -> Weight -> BN ->

src/models.py:144

↓ 1 callersFunctionbatch_global_rigid_transformation

Computes absolute joint locations given pose. rotate_base: if True, rotates the global rotation by 90 deg in x axis. if False, this is t

src/tf_smpl/batch_lbs.py:133

↓ 1 callersFunctionbatch_orth_proj_optcam

Solves for best sale and translation in 2D, i.e. gives (s, t) such that ||s(x + t) - x_gt||^2 X is N x K x 2, for [x, y] pred (via identi

src/tf_smpl/projection.py:32

↓ 1 callersFunctionbatch_rot2aa

Rs is B x 3 x 3 void cMathUtil::RotMatToAxisAngle(const tMatrix& mat, tVector& out_axis, double& out_th

src/tf_smpl/batch_lbs.py:63

↓ 1 callersFunctionbatch_skew

vec is N x 3, batch_size is int returns N x 3 x 3. Skew_sym version of each matrix.

src/tf_smpl/batch_lbs.py:15

↓ 1 callersMethodbuild_hmr_model

(self)

src/trainer_sequence_fc.py:490

↓ 1 callersMethodbuild_model

Pseudo code: Omega_mu = get_mean_omega() - this has been folded into init though h_0 = initialize to 0 for t in [1:T]

src/trainer_sequence_fc.py:551

↓ 1 callersMethodbuild_model

(self)

src/datasets/resnet_extractor.py:40

↓ 1 callersMethodbuild_model

(self)

src/util/tube_augmentation.py:258

↓ 1 callersMethodbuild_test_model

(self)

src/evaluation/tester.py:169

↓ 1 callersFunctioncall_hmr_ief

Wrapper for doing HMR-style IEF. If predict_delta, then also makes num_delta_t predictions forward and backward in time, with each step

src/models.py:299

↓ 1 callersFunctioncall_hmr_ief

Wrapper for doing HMR-style IEF. If predict_delta, then also makes num_delta_t predictions forward and backward in time, with each step

src/ops.py:184

↓ 1 callersFunctioncompute_accel

Computes acceleration of 3D joints. Args: joints (Nx25x3). Returns: Accelerations (N-2).

src/evaluation/eval_util.py:14

↓ 1 callersMethodcompute_all_smpl

(cls)

src/omega.py:339

↓ 1 callersFunctioncompute_delta_poses

Computes the delta poses for a single video, sampled at a frame skip of 1 to max_subsample_rate. Args: poses (Nx72): Array of ax

src/datasets/smpl_to_tfrecords.py:95

↓ 1 callersFunctioncompute_deltas_batched

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

src/util/tf_ops.py:11

↓ 1 callersFunctioncompute_error_3d

Returns MPJPE after pelvis alignment and MPJPE after Procrustes. Should evaluate only on the 14 common joints. Args: gt3ds (Nx14

src/evaluation/eval_util.py:30

↓ 1 callersFunctioncompute_error_accel

Computes acceleration error: 1/(n-2) \sum_{i=1}^{n-1} X_{i-1} - 2X_i + X_{i+1} Note that for each frame that is not visible, three e

src/evaluation/eval_util.py:63

↓ 1 callersFunctioncompute_error_kp

Compute the keypoint error (mean difference in pixels), keypoint error after Procrustes Analysis, and percent correct keypoints. Args:

src/evaluation/eval_util.py:97

↓ 1 callersFunctioncompute_joints

Computes joints for gendered

src/datasets/threedpw/compute_neutral_shape.py:162

↓ 1 callersFunctioncompute_loss_d_fake

(out_fake)

src/ops.py:131

↓ 1 callersFunctioncompute_loss_d_real

(out_real)

src/ops.py:135

↓ 1 callersFunctioncompute_loss_e_fake

(out_fake)

src/ops.py:127

↓ 1 callersFunctioncompute_loss_e_kp_optcam

computes optimal scale s and translation t=[tx; ty], given kp_gt and kp_pred. then computes: L_{KP}=\sum_{i=1}^K v_i||x_i - s*(\hat{x_i}

src/ops.py:14

↓ 1 callersFunctioncompute_loss_e_smooth

Args: joints_prev (NxKx3): Predicted joint locations for time t - 1. joints_curr (NxKx3): Predicted joint locations for time t.

src/ops.py:113

↓ 1 callersFunctioncompute_loss_shape

L2 loss on shapes.

src/ops.py:160

↓ 1 callersMethodcompute_losses_batched

(self)

src/trainer_sequence_fc.py:791

↓ 1 callersMethodcompute_losses_static

(self)

src/trainer_sequence_fc.py:955

↓ 1 callersFunctioncompute_opt_cam_with_vis

Computes the optimal camera [scale, tx, ty] to map 2D keypoints got to 2D keypoints want with boolean visibility indicator vis.

src/evaluation/eval_util.py:235

↓ 1 callersMethodcompute_phis

Runs forward pass of the model. Args: images (BxHxWx3). Returns: phis (Bx2048).

src/datasets/resnet_extractor.py:58

↓ 1 callersFunctioncompute_similarity_transform

Computes a similarity transform (sR, t) that takes a set of 3D points S1 (3 x N) closest to a set of 3D points S2, where R is an 3x3 rotation

src/evaluation/eval_util.py:177

↓ 1 callersFunctionconvert_to_example

Build an Example proto for an image example. Args: pose (72). shape (10). Returns: Example proto.

src/datasets/smpl_to_tfrecords.py:53

↓ 1 callersFunctionconvert_to_example_temporal

Builds example proto for a changes in pose. Args: pose (50x24x3): Axis-angle representation of the current pose. Returns:

src/datasets/smpl_to_tfrecords.py:78

↓ 1 callersFunctiondecode_jpeg

Decode a JPEG string into one 3-D float image Tensor. Args: image_buffer: scalar string Tensor. name: Optional name for name_sco

src/util/data_utils.py:458

↓ 1 callersFunctiondraw_text

content is a dict. draws key: val on image Assumes key is str, val is float

src/util/render/render_utils.py:9

↓ 1 callersFunctiondump_frames

Extracts all frames from the video at vid_path and saves them inside of out_dir.

extract_tracks.py:42

↓ 1 callersMethodencode_png

(self, image)

src/datasets/common.py:69

↓ 1 callersMethodgather_losses

Consolidates all loss functions for encoder and discriminator optimizers and setups summaries.

src/trainer_sequence_fc.py:700

↓ 1 callersFunctionget_3dpw2coco

()

src/datasets/threedpw/read_3dpw.py:5

↓ 1 callersFunctionget_all_bbox_params

Finds bounding box parameters for all keypoints. Look for sequences in the middle with no predictions and linearly interpolate the bbox

src/util/smooth_bbox.py:64

↓ 1 callersFunctionget_all_data

(img_dir, split='train')

src/datasets/h36_to_tfrecords_video.py:294

↓ 1 callersFunctionget_eval_path_name

Gets path name to store computed evaluations.

src/evaluation/prediction.py:83

↓ 1 callersFunctionget_labels_poseflow

Returns the poses for each person tracklet. Each pose has dimension num_kp x 3 (x,y,vis) if the person is visible in the current frame.

demo_video.py:61

↓ 1 callersMethodget_loader

Returns: batch_dict (dict): image (BxTxHxWx3). label (BxTx19x3). pose (BxTx24x3x3

src/data_loader_sequence.py:60

↓ 1 callersMethodget_output

Gets discriminator's predictions for each pose and all poses. Args: poses (Nx23x1x9). Returns: Pred

src/discriminators.py:15

↓ 1 callersFunctionget_predictions

If predictions exist, load from pickle. Otherwise, makes the predictions.

src/evaluation/prediction.py:119

↓ 1 callersMethodget_raw

Returns: Raw Omega (BxTx85).

src/omega.py:331

↓ 1 callersFunctionget_result_path_name

Gets path name to store computed evaluations.

src/evaluation/prediction.py:64

↓ 1 callersFunctionget_seq_data

Args: anno_pkl (str): Path to annotation directory. img_dir (str): Path to image directory. Returns: im_paths: List

src/datasets/3dpw_to_tfrecords_video.py:43

↓ 1 callersFunctionget_seq_labels

(data_dir)

src/datasets/upenn_to_tfrecords_video.py:397

↓ 1 callersMethodget_smpl_loader

Loads real pose and shape parameters from mocap data. Returns: pose_batch (BxTx216). shape_batch (BxTx10).

src/data_loader_sequence.py:159

↓ 1 callersMethodget_unfrozen_E_vars

(self)

src/trainer_sequence_fc.py:681

↓ 1 callersMethodget_vars

(self)

src/discriminators.py:80

↓ 1 callersFunctionkp_to_bbox_param

Finds the bounding box parameters from the 2D keypoints. Args: kp (Kx3): 2D Keypoints. vis_thresh (float): Threshold for vis

src/util/smooth_bbox.py:37

↓ 1 callersMethodload_mean_omega

(self)

src/trainer_sequence_fc.py:447

↓ 1 callersMethodload_mean_params

(self)

src/evaluation/tester.py:118

↓ 1 callersFunctionmain

(model)

demo_video.py:220

↓ 1 callersFunctionmain

(opts)

extract_tracks.py:153

↓ 1 callersFunctionmain

(config)

src/main.py:14

↓ 1 callersFunctionmain

(raw_data_root, output_root, frame_skip)

src/datasets/h36/read_human36m.py:309

↓ 1 callersFunctionmain

(config)

src/evaluation/eval.py:353

↓ 1 callersMethodmake_alpha

(self, rend, mask)

src/util/render/nmr_renderer.py:227

↓ 1 callersFunctionpick_sequences

Returns random subset of length T. Args: N: Total number of samples. T: Desired sequence length. Returns: Tenso

src/util/data_utils.py:340

↓ 1 callersFunctionplot_points

(points_, img, points0_=None, radius=10)

src/datasets/h36/read_human36m.py:67

↓ 1 callersMethodpng_to_jpeg

(self, image_data)

src/datasets/common.py:47

↓ 1 callersMethodpredict

Runs forward pass of model. Args: images (BxTxHxWx3): Images to predict. Returns: dict.

src/evaluation/tester.py:229

↓ 1 callersMethodpredict_fc

f_temporal_enc gets the image features and then calls the appropriate model.

src/trainer_sequence_fc.py:635

↓ 1 callersFunctionpredict_on_tracks

(model, img_dir, poseflow_path, output_path, track_id, trim_length)

demo_video.py:124

↓ 1 callersMethodprepare

(self)

src/datasets/resnet_extractor.py:54

↓ 1 callersMethodprepare

Restores variables from checkpoint. Args: resnet_path (str): Optional path to load resnet weights.

src/evaluation/tester.py:92

↓ 1 callersFunctionprepare_dirs

(config, prefix=[])

src/config.py:152

↓ 1 callersMethodpreprocess_image

(self, image, image_size, label, center, pose, gt3d, trans, scale, rotate, flip)

src/util/tube_augmentation.py:114

↓ 1 callersFunctionprocess_3dpw

(data_dir, out_dir, visualize, split)

src/datasets/3dpw_to_tfrecords_video.py:118

↓ 1 callersFunctionprocess_h36

(data_dir, out_dir, split='val', feature_extractor=None,

src/datasets/h36_to_tfrecords_video.py:386

↓ 1 callersFunctionprocess_image

(im_path, gt2d, coder, DRAW=False)

src/datasets/upenn_to_tfrecords_video.py:80

↓ 1 callersFunctionprocess_image

(im_path, gt2d, coder, bbox_param, DRAW=False)

src/datasets/insta_variety_to_tfrecords.py:113

↓ 1 callersFunctionprocess_image

(im_path, gt2d, coder, bbox_param, DRAW=False)

src/datasets/video_in_the_wild_to_tfrecords.py:114

↓ 1 callersFunctionprocess_image

( im_path, gt2d, cam, coder, pose=None, shape=None, gt

src/datasets/h36_to_tfrecords_video.py:87

↓ 1 callersFunctionprocess_image

Processes an image, producing 224x224 crop. Args: im_path (str). gt2d (19x3). coder (tf.ImageCoder). bbox_pa

src/datasets/make_test_tfrecords.py:164

↓ 1 callersFunctionprocess_smpl_mocap

(all_pkls, out_dir, num_shards, dataset_name)

src/datasets/smpl_to_tfrecords.py:120

↓ 1 callersFunctionprocess_smpl_mocap_temporal

(all_pkls, out_dir, num_shards, dataset_name)

src/datasets/smpl_to_tfrecords.py:161

↓ 1 callersFunctionprocess_videos

Used for training. seq_path is a path to the directory of a seq. extracts all kps and images, shuffles and then stores in a shard.

src/datasets/upenn_to_tfrecords_video.py:322

↓ 1 callersFunctionprocess_videos

(out_dir, seq_paths, all_kps_raw, feature_extractor=N

src/datasets/insta_variety_to_tfrecords.py:366

↓ 1 callersFunctionprocess_videos

(out_dir, seq_paths, all_kps_raw, feature_extractor=N

src/datasets/video_in_the_wild_to_tfrecords.py:367

↓ 1 callersFunctionprocess_videos_test

seq_path is a path to the directory of a seq. each video becomes a tfrecord.

src/datasets/insta_variety_to_tfrecords.py:417

↓ 1 callersFunctionprocess_videos_test

seq_path is a path to the directory of a seq. each video becomes a tfrecord.

src/datasets/video_in_the_wild_to_tfrecords.py:418

↓ 1 callersFunctionprocrustes2d_vis

Solves for the optimal sale and translation in 2D, i.e. gives (s, t) such that ||s(x + t) - x_gt||^2 on *visible* points. Gradient i

src/tf_smpl/projection.py:48

↓ 1 callersFunctionproject

(X, c)

src/datasets/h36_to_tfrecords_video.py:165

↓ 1 callersFunctionquat_rotate

Rotate points by quaternions. Args: X: B X N X 3 points q: B X 4 quaternions Returns: X_rot: B X N X 3 (rotated poin

src/util/render/torch_utils.py:55

↓ 1 callersFunctionread_action_name

(xml_path, sbj_id, actionno, trialno)

src/datasets/h36/read_human36m.py:171

↓ 1 callersFunctionread_cam_parameters

(xml_path, sbj_id, cam_id)

src/datasets/h36/read_human36m.py:131

↓ 1 callersFunctionread_frames

(path, n_frames=None)

src/datasets/h36/read_human36m.py:200

↓ 1 callersFunctionread_upenn

(base_dir)

src/datasets/upenn/read_upenn.py:150

↓ 1 callersMethodrecord_loss_proportions

Prints out loss proportions and writes it to a file. Args: loss_proportions (dict): Dictionary of loss values.

src/trainer_sequence_fc.py:1319

↓ 1 callersFunctionrectify_joints

(joints, camR)

src/datasets/3dpw_to_tfrecords_video.py:95

Functions299 in github.com/akanazawa/human_dynamics