thanks to mingyuan-zhang ❤

data/database/generate_kit.py

@@ -0,0 +1,63 @@
+import os
+import torch
+import numpy as np
+import clip
+from tqdm import tqdm
+
+device = 'cpu'
+clip_model, _ = clip.load('ViT-B/32', device)
+
+data_root_dir = "../datasets/kit_ml"
+data_clip_dir = os.path.join(data_root_dir, "clip_feats")
+data_caption_dir = os.path.join(data_root_dir, "texts")
+data_motion_dir = os.path.join(data_root_dir, "motions")
+train_split = os.path.join(data_root_dir, "train.txt")
+
+all_text_features = []
+all_captions = []
+all_motions = []
+all_m_lengths = []
+all_clip_seq_features = []
+
+std = np.load(os.path.join(data_root_dir, "std.npy"))
+mean = np.load(os.path.join(data_root_dir, "mean.npy"))
+
+for filename in tqdm(open(train_split)):
+    filename = filename.strip()
+    caption_file = os.path.join(data_caption_dir, filename + ".txt")
+    caption = open(caption_file).readlines()[0].strip()
+    text = clip.tokenize([caption], truncate=True).to(device)
+    with torch.no_grad():
+        text_feature = clip_model.encode_text(text)[0].numpy()
+    all_text_features.append(text_feature)
+    all_captions.append(caption)
+    motion_file = os.path.join(data_motion_dir, filename + ".npy")
+    motion_data = np.load(motion_file)
+    # import pdb; pdb.set_trace()
+    motion_data = (motion_data - mean) / (std + 1e-9)
+    motion_data = motion_data[:196]
+    motion = np.zeros((196, 251))
+    motion[:motion_data.shape[0], :] = motion_data
+    all_motions.append(motion)
+    m_length = motion_data.shape[0]
+    all_m_lengths.append(m_length)
+    clip_feat_file = os.path.join(data_clip_dir, filename + ".npy")
+    clip_feat = np.load(clip_feat_file)[0]
+    all_clip_seq_features.append(clip_feat)
+
+all_text_features = np.array(all_text_features)
+all_captions = np.array(all_captions)
+all_motions = np.array(all_motions)
+all_m_lengths = np.array(all_m_lengths)
+all_clip_seq_features = np.array(all_clip_seq_features)
+
+output = {
+    'text_features': all_text_features,
+    'captions': all_captions,
+    'motions': all_motions,
+    'm_lengths': all_m_lengths,
+    'clip_seq_features': all_clip_seq_features
+}
+
+npz_path = "kit_text_train.npz"
+np.savez_compressed(npz_path, **output)

data/database/generate_t2m.py

@@ -0,0 +1,63 @@
+import os
+import torch
+import numpy as np
+import clip
+from tqdm import tqdm
+
+device = 'cpu'
+clip_model, _ = clip.load('ViT-B/32', device)
+
+data_root_dir = "../datasets/human_ml3d"
+data_clip_dir = os.path.join(data_root_dir, "clip_feats")
+data_caption_dir = os.path.join(data_root_dir, "texts")
+data_motion_dir = os.path.join(data_root_dir, "motions")
+train_split = os.path.join(data_root_dir, "train.txt")
+
+all_text_features = []
+all_captions = []
+all_motions = []
+all_m_lengths = []
+all_clip_seq_features = []
+
+std = np.load(os.path.join(data_root_dir, "std.npy"))
+mean = np.load(os.path.join(data_root_dir, "mean.npy"))
+
+for filename in tqdm(open(train_split)):
+    filename = filename.strip()
+    caption_file = os.path.join(data_caption_dir, filename + ".txt")
+    caption = open(caption_file).readlines()[0].strip()
+    text = clip.tokenize([caption], truncate=True).to(device)
+    with torch.no_grad():
+        text_feature = clip_model.encode_text(text)[0].numpy()
+    all_text_features.append(text_feature)
+    all_captions.append(caption)
+    motion_file = os.path.join(data_motion_dir, filename + ".npy")
+    motion_data = np.load(motion_file)
+    # import pdb; pdb.set_trace()
+    motion_data = (motion_data - mean) / (std + 1e-9)
+    motion_data = motion_data[:196]
+    motion = np.zeros((196, 263))
+    motion[:motion_data.shape[0], :] = motion_data
+    all_motions.append(motion)
+    m_length = motion_data.shape[0]
+    all_m_lengths.append(m_length)
+    clip_feat_file = os.path.join(data_clip_dir, filename + ".npy")
+    clip_feat = np.load(clip_feat_file)[0]
+    all_clip_seq_features.append(clip_feat)
+
+all_text_features = np.array(all_text_features)
+all_captions = np.array(all_captions)
+all_motions = np.array(all_motions)
+all_m_lengths = np.array(all_m_lengths)
+all_clip_seq_features = np.array(all_clip_seq_features)
+
+output = {
+    'text_features': all_text_features,
+    'captions': all_captions,
+    'motions': all_motions,
+    'm_lengths': all_m_lengths,
+    'clip_seq_features': all_clip_seq_features
+}
+
+npz_path = "t2m_text_train.npz"
+np.savez_compressed(npz_path, **output)

data/database/kit_text_train.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6cb18cdf63734d9e34c2abf402269d7a4aaea02c727c894cb7e0161de10fe053
+size 962043519

data/database/t2m_text_train.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e597ca6ab524a73f20074d8e7f0fa013d99872e5b590e778aecaefcee1172b77
+size 5852293680

data/datasets/human_ml3d.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:77b13f18bbd01f6d052b70b5fa27a11fc474aa58107dd5f97f687bd991a71017
+size 10397657625

data/datasets/kit_ml.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1475d2c8b54ea110c7b731ec63f06bcc0dac0751eda5a551b9bc3c54d6b7666d
+size 1326887901

data/evaluators/human_ml3d/finest.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:881e45bea6fb84b15eca6e7cdd75d602b58a72f2038c223ee96ab75d4f684ada
+size 245580211

data/evaluators/kit_ml/finest.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f786d733ca1bfa5dc1474554f167e1f608ec629868061e51ccfdb812ea61e6a
+size 245481907

data/glove/our_vab_data.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:73ed9b73ee5bc5ab683661c132b9f98fe305672dffdf70ceeac1bc447543d668
+size 10077728

data/glove/our_vab_idx.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d85dce83d2c27a92bb94bb51dc1f55a04fcaa328c359d23eccf648e296c16493
+size 79811

data/glove/our_vab_words.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fac88da1be6f00d36b72b88c9745c8ac35a94a7bd6ccf1fa3bab380faf8c2e0d
+size 67470

logs/finemogen/finemogen_kit/latest.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:66b7ea74dca87a1940ad29205b66032edeba0fcf62ef4a007ed3764f415b4c1d
+size 261983599

logs/finemogen/finemogen_t2m/latest.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8fd39808c1f0a9f57374e76ec99388359bd639a85273c5d8bd6f93f2097e0749
+size 261996143

logs/mdm/mdm_t2m/latest.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8810255fb8df9eed6211537de9826f07ff73862f367cbf91532d84fd4c9a497e
+size 81791550

logs/motiondiffuse/motiondiffuse_kit/latest.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4056909b968615df88a8a81d1566dcbe2de499ac816832181ceeac2a453aedf3
+size 953909508

logs/motiondiffuse/motiondiffuse_t2m/latest.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc5cc2269fb42237910e75b2d04357ccc25d8ad59ae5d27d8a1134e8f68e3860
+size 953958724

logs/remodiffuse/remodiffuse_kit/latest.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7da46a5ac84be587c715325ac33f2c837060ee6b7d212876406678d7f1c3c64b
+size 1168873949

logs/remodiffuse/remodiffuse_t2m/latest.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d36ca2f5372a0aa8a4be054e61af9a36529254ad94944e440e8a8e7dfe5e8327
+size 1169095901