Upload 6 files

_DATA/data/mano/MANO_RIGHT.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:45d60aa3b27ef9107a7afd4e00808f307fd91111e1cfa35afd5c4a62de264767
+size 3821356

_DATA/data/mano_mean_params.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efc0ec58e4a5cef78f3abfb4e8f91623b8950be9eff8b8e0dbb0d036ebc63988
+size 1178

_DATA/hamer_ckpts/checkpoints/hamer.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5cc06f294d88a92dee24e603480aab04de532b49f0e08200804ee7d90e16f53
+size 2689536166

_DATA/hamer_ckpts/dataset_config.yaml

@@ -0,0 +1,42 @@
+COCOW-TRAIN:
+  TYPE: ImageDataset
+  URLS: hamer_training_data/dataset_tars/cocow-train/{000000..000036}.tar
+  epoch_size: 78666
+DEX-TRAIN:
+  TYPE: ImageDataset
+  URLS: hamer_training_data/dataset_tars/dex-train/{000000..000406}.tar
+  epoch_size: 406888
+FREIHAND-MOCAP:
+  DATASET_FILE: hamer_training_data/freihand_mocap.npz
+FREIHAND-TRAIN:
+  TYPE: ImageDataset
+  URLS: hamer_training_data/dataset_tars/freihand-train/{000000..000130}.tar
+  epoch_size: 130240
+H2O3D-TRAIN:
+  TYPE: ImageDataset
+  URLS: hamer_training_data/dataset_tars/h2o3d-train/{000000..000060}.tar
+  epoch_size: 121996
+HALPE-TRAIN:
+  TYPE: ImageDataset
+  URLS: hamer_training_data/dataset_tars/halpe-train/{000000..000022}.tar
+  epoch_size: 34289
+HO3D-TRAIN:
+  TYPE: ImageDataset
+  URLS: hamer_training_data/dataset_tars/ho3d-train/{000000..000083}.tar
+  epoch_size: 83325
+INTERHAND26M-TRAIN:
+  TYPE: ImageDataset
+  URLS: hamer_training_data/dataset_tars/interhand26m-train/{000000..001056}.tar
+  epoch_size: 1424632
+MPIINZSL-TRAIN:
+  TYPE: ImageDataset
+  URLS: hamer_training_data/dataset_tars/mpiinzsl-train/{000000..000015}.tar
+  epoch_size: 15184
+MTC-TRAIN:
+  TYPE: ImageDataset
+  URLS: hamer_training_data/dataset_tars/mtc-train/{000000..000306}.tar
+  epoch_size: 363947
+RHD-TRAIN:
+  TYPE: ImageDataset
+  URLS: hamer_training_data/dataset_tars/rhd-train/{000000..000041}.tar
+  epoch_size: 61705

_DATA/hamer_ckpts/model_config.yaml

@@ -0,0 +1,111 @@
+task_name: train
+tags:
+- dev
+train: true
+test: false
+ckpt_path: null
+seed: null
+DATASETS:
+  TRAIN:
+    FREIHAND-TRAIN:
+      WEIGHT: 0.25
+    INTERHAND26M-TRAIN:
+      WEIGHT: 0.25
+    MTC-TRAIN:
+      WEIGHT: 0.1
+    RHD-TRAIN:
+      WEIGHT: 0.05
+    COCOW-TRAIN:
+      WEIGHT: 0.1
+    HALPE-TRAIN:
+      WEIGHT: 0.05
+    MPIINZSL-TRAIN:
+      WEIGHT: 0.05
+    HO3D-TRAIN:
+      WEIGHT: 0.05
+    H2O3D-TRAIN:
+      WEIGHT: 0.05
+    DEX-TRAIN:
+      WEIGHT: 0.05
+  VAL:
+    FREIHAND-TRAIN:
+      WEIGHT: 1.0
+  MOCAP: FREIHAND-MOCAP
+  BETAS_REG: true
+  CONFIG:
+    SCALE_FACTOR: 0.3
+    ROT_FACTOR: 30
+    TRANS_FACTOR: 0.02
+    COLOR_SCALE: 0.2
+    ROT_AUG_RATE: 0.6
+    TRANS_AUG_RATE: 0.5
+    DO_FLIP: false
+    FLIP_AUG_RATE: 0.0
+    EXTREME_CROP_AUG_RATE: 0.0
+    EXTREME_CROP_AUG_LEVEL: 1
+extras:
+  ignore_warnings: false
+  enforce_tags: true
+  print_config: true
+exp_name: hamer
+MANO:
+  DATA_DIR: _DATA/data/
+  MODEL_PATH: data/mano
+  GENDER: neutral
+  NUM_HAND_JOINTS: 15
+  MEAN_PARAMS: data/mano_mean_params.npz
+  CREATE_BODY_POSE: false
+EXTRA:
+  FOCAL_LENGTH: 5000
+  NUM_LOG_IMAGES: 4
+  NUM_LOG_SAMPLES_PER_IMAGE: 8
+  PELVIS_IND: 0
+GENERAL:
+  TOTAL_STEPS: 1000000
+  LOG_STEPS: 1000
+  VAL_STEPS: 1000
+  CHECKPOINT_STEPS: 10000
+  CHECKPOINT_SAVE_TOP_K: 1
+  NUM_WORKERS: 8
+  PREFETCH_FACTOR: 2
+TRAIN:
+  LR: 1.0e-05
+  WEIGHT_DECAY: 0.0001
+  BATCH_SIZE: 32
+  LOSS_REDUCTION: mean
+  NUM_TRAIN_SAMPLES: 2
+  NUM_TEST_SAMPLES: 64
+  POSE_2D_NOISE_RATIO: 0.01
+  SMPL_PARAM_NOISE_RATIO: 0.005
+MODEL:
+  IMAGE_SIZE: 256
+  IMAGE_MEAN:
+  - 0.485
+  - 0.456
+  - 0.406
+  IMAGE_STD:
+  - 0.229
+  - 0.224
+  - 0.225
+  BACKBONE:
+    TYPE: vit
+    PRETRAINED_WEIGHTS: hamer_training_data/vitpose_backbone.pth
+  MANO_HEAD:
+    TYPE: transformer_decoder
+    IN_CHANNELS: 2048
+    TRANSFORMER_DECODER:
+      depth: 6
+      heads: 8
+      mlp_dim: 1024
+      dim_head: 64
+      dropout: 0.0
+      emb_dropout: 0.0
+      norm: layer
+      context_dim: 1280
+LOSS_WEIGHTS:
+  KEYPOINTS_3D: 0.05
+  KEYPOINTS_2D: 0.01
+  GLOBAL_ORIENT: 0.001
+  HAND_POSE: 0.001
+  BETAS: 0.0005
+  ADVERSARIAL: 0.0005

_DATA/vitpose_ckpts/vitpose+_huge/wholebody.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0555e1e2392e6a2be2d9265368f344d70ccbfd656ad480aa5c1de2e604519c9
+size 3807742341