evn

app.py

@@ -38,13 +38,14 @@ def predict(file_path: str):
     temp_bash_file = create_bash_file(temp_file_path)
     
     os.system(f"bash {temp_bash_file}")
+    return temp_file_path
     
 
 demo = gr.Interface(
     predict,
     # gr.Dataframe(type="numpy", datatype="number", row_count=5, col_count=3),
     gr.File(type="filepath"),
-    "numpy",
+    "file",
     cache_examples=False
 )
 

data_loaders/humanml/data/dataset_ours_single_seq.py

@@ -3451,7 +3451,8 @@ class GRAB_Dataset_V19(torch.utils.data.Dataset): # GRAB datasset and
             base_pts = object_pc_th[:, base_pts_mask] # ws x nn_valid_obj_pcs x 3 #
             base_normals = object_normal_th[:, base_pts_mask] # ws x nn_valid_obj_pcs x 3 #
             nn_base_pts = self.nn_base_pts
-            base_pts_idxes = utils.farthest_point_sampling(base_pts[0:1], n_sampling=nn_base_pts)
+            if base_pts.size(1) > nn_base_pts:
+                base_pts_idxes = utils.farthest_point_sampling(base_pts[0:1], n_sampling=nn_base_pts)
             base_pts_idxes = base_pts_idxes[:nn_base_pts]
             base_pts = base_pts[:, base_pts_idxes]
             base_normals = base_normals[:, base_pts_idxes]
@@ -7720,10 +7721,11 @@ class GRAB_Dataset_V19_HHO(torch.utils.data.Dataset): # GRAB datasset #
         base_pts = object_pc_th[:, base_pts_mask] # ws x nn_valid_obj_pcs x 3 #
         base_normals = object_normal_th[:, base_pts_mask] # ws x nn_valid_obj_pcs x 3 #
         nn_base_pts = self.nn_base_pts
-        base_pts_idxes = utils.farthest_point_sampling(base_pts[0:1], n_sampling=nn_base_pts)
-        base_pts_idxes = base_pts_idxes[:nn_base_pts]
-        base_pts = base_pts[:, base_pts_idxes]
-        base_normals = base_normals[:, base_pts_idxes]
+        if base_pts.size(-1) > nn_base_pts:
+            base_pts_idxes = utils.farthest_point_sampling(base_pts[0:1], n_sampling=nn_base_pts)
+            base_pts_idxes = base_pts_idxes[:nn_base_pts]
+            base_pts = base_pts[:, base_pts_idxes]
+            base_normals = base_normals[:, base_pts_idxes]
         
         base_pts_global_orient_mtx = object_global_orient_mtx_th # ws x 3 x 3 #
         base_pts_transl = object_trcansl_th # ws x 3 # 

requirements.txt

@@ -2,11 +2,11 @@
 -f https://data.pyg.org/whl/torch-1.12.1+cpu.html
 # pip==20.2.4
 torch==1.12.1
-torchvision==0.13.1
-torchaudio==0.12.1
+# torchvision==0.13.1
+# torchaudio==0.12.1
 blobfile==2.0.1
 manopth @ git+https://github.com/hassony2/manopth.git
-numpy
+numpy==1.23.1
 psutil
 scikit-learn
 scipy
@@ -14,3 +14,6 @@ tensorboard
 tensorboardx
 tqdm
 trimesh
+clip
+chumpy
+opencv-python

sample/reconstruct_data.py

@@ -4,19 +4,19 @@ sys.path.insert(0, '..')
 
 import torch
 import torch.nn.functional as F
-from torch import optim, nn
+from torch import optim
 import numpy as np
-import os, argparse, copy, json
-import pickle as pkl
-from scipy.spatial.transform import Rotation as R
-from psbody.mesh import Mesh
+# import os, argparse, copy, json
+# import pickle as pkl
+# from scipy.spatial.transform import Rotation as R
+# from psbody.mesh import Mesh
 from manopth.manolayer import ManoLayer
 # from dataloading import GRAB_Single_Frame, GRAB_Single_Frame_V6, GRAB_Single_Frame_V7, GRAB_Single_Frame_V8, GRAB_Single_Frame_V9, GRAB_Single_Frame_V9_Ours, GRAB_Single_Frame_V10
 # use_trans_encoders
 # from model import TemporalPointAE, TemporalPointAEV2, TemporalPointAEV5, TemporalPointAEV6, TemporalPointAEV7, TemporalPointAEV8, TemporalPointAEV9, TemporalPointAEV10, TemporalPointAEV4, TemporalPointAEV3_Real, TemporalPointAEV11, TemporalPointAEV12, TemporalPointAEV13, TemporalPointAEV14, TemporalPointAEV17, TemporalPointAEV19, TemporalPointAEV20, TemporalPointAEV21, TemporalPointAEV22, TemporalPointAEV23, TemporalPointAEV24, TemporalPointAEV25, TemporalPointAEV26
-import trimesh
+# import trimesh
 from utils import *
-import utils
+# import utils
 import utils.model_util as model_util
 # from anchorutils import anchor_load_driver, recover_anchor, recover_anchor_batch
 

train/training_loop_ours.py

@@ -57,7 +57,7 @@ class TrainLoop:
         self.num_steps = args.num_steps
         self.num_epochs = self.num_steps // len(self.data) + 1
 
-        self.sync_cuda = torch.cuda.is_available()
+        self.sync_cuda = False # torch.cuda.is_available()
 
         if self.args.finetune_with_cond: # finetune_with_cond -> 
             self._load_and_sync_parameters_cond() # load parameters here 
@@ -85,10 +85,10 @@ class TrainLoop:
             # Model was resumed, either due to a restart or a checkpoint
             # being specified at the command line.
 
-        print(f"dist_utils: {dist_util.dev()}")
+        print(f"dist_utils: {dist_util.dev()}") 
         self.device = torch.device("cpu")
-        if torch.cuda.is_available() and dist_util.dev() != 'cpu':
-            self.device = torch.device(dist_util.dev())
+        # if torch.cuda.is_available() and dist_util.dev() != 'cpu':
+        #     self.device = torch.device(dist_util.dev())
 
         self.schedule_sampler_type = 'uniform'
         self.schedule_sampler = create_named_schedule_sampler(self.schedule_sampler_type, diffusion)

utils/dist_util.py

@@ -45,10 +45,12 @@ def dev():
     """
     Get the device to use for torch.distributed.
     """
+    
     global used_device
-    if th.cuda.is_available() and used_device>=0:
-        return th.device(f"cuda:{used_device}")
     return th.device("cpu")
+    # if th.cuda.is_available() and used_device>=0:
+    #     return th.device(f"cuda:{used_device}")
+    # return th.device("cpu")
 
 
 def load_state_dict(path, **kwargs):