update CP config

README.md

@@ -1,6 +1,6 @@
 # TensoRF
 ## [Project page](https://apchenstu.github.io/TensoRF/) | [Paper](https://arxiv.org/abs/2203.09517)
-This repository contains a pytorch implementation for the paper: [TensoRF: Tensorial Radiance Fields](https://arxiv.org/abs/2103.15595). Our work present a novel approach to model and reconstruct radiance fields, which achieves super
+This repository contains a pytorch implementation for the paper: [TensoRF: Tensorial Radiance Fields](https://arxiv.org/abs/2203.09517). Our work present a novel approach to model and reconstruct radiance fields, which achieves super
 **fast** training process, **compact** memory footprint and **state-of-the-art** rendering quality.<br><br>
 
 
@@ -36,11 +36,12 @@ python train.py --config configs/lego.txt
 
 we provide a few examples in the configuration folder, please note:
 
- `dataset_name`, choices = ['blender', 'llff', 'nsvf', 'dtu','tankstemple'];
+ `dataset_name`, choices = ['blender', 'llff', 'nsvf', 'tankstemple'];
 
  `shadingMode`, choices = ['MLP_Fea', 'SH'];
 
- `model_name`, choices = ['TensorVMSplit', 'TensorCP'], corresponding to the VM and CP decomposition;
+ `model_name`, choices = ['TensorVMSplit', 'TensorCP'], corresponding to the VM and CP decomposition. 
+ You need to uncomment the last a few rows of the configuration file if you want to training with the TensorCP model；
 
  `n_lamb_sigma` and `n_lamb_sh` are string type refer to the basis number of density and appearance along XYZ
 dimension;

configs/lego.txt

@@ -1,7 +1,7 @@
 
 dataset_name = blender
 datadir = ./data/nerf_synthetic/lego
-expname = tensorf_lego
+expname = tensorf_lego_CP
 basedir = ./log
 
 n_iters = 30000
@@ -19,6 +19,8 @@ render_test = 1
 
 n_lamb_sigma = [16,16,16]
 n_lamb_sh = [48,48,48]
+model_name = TensorVMSplit
+
 
 shadingMode = MLP_Fea
 fea2denseAct = softplus
@@ -29,3 +31,11 @@ fea_pe = 2
 L1_weight_inital = 8e-5
 L1_weight_rest = 4e-5
 rm_weight_mask_thre = 1e-4
+
+## please uncomment following configuration if hope to training on cp model
+#model_name = TensorCP
+#n_lamb_sigma = [96]
+#n_lamb_sh = [288]
+#N_voxel_final = 125000000 # 500**3
+#L1_weight_inital = 1e-5
+#L1_weight_rest = 1e-5

configs/truck.txt

@@ -30,3 +30,11 @@ fea_pe = 2
 TV_weight_density = 0.1
 TV_weight_app = 0.01
 
+## please uncomment following configuration if hope to training on cp model
+#model_name = TensorCP
+#n_lamb_sigma = [96]
+#n_lamb_sh = [288]
+#N_voxel_final = 125000000 # 500**3
+#L1_weight_inital = 1e-5
+#L1_weight_rest = 1e-5
+

configs/wineholder.txt

@@ -29,3 +29,11 @@ fea_pe = 2
 L1_weight_inital = 8e-5
 L1_weight_rest = 4e-5
 rm_weight_mask_thre = 1e-4
+
+## please uncomment following configuration if hope to training on cp model
+#model_name = TensorCP
+#n_lamb_sigma = [96]
+#n_lamb_sh = [288]
+#N_voxel_final = 125000000 # 500**3
+#L1_weight_inital = 1e-5
+#L1_weight_rest = 1e-5

models/tensoRF.py

@@ -303,16 +303,14 @@ def shrink(self, new_aabb):
  self.update_stepSize((newSize[0], newSize[1], newSize[2]))
 
 
-
-
 class TensorCP(TensorBase):
  def __init__(self, aabb, gridSize, device, **kargs):
  super(TensorCP, self).__init__(aabb, gridSize, device, **kargs)
 
 
  def init_svd_volume(self, res, device):
- self.density_line = self.init_one_svd(self.density_n_comp, self.gridSize, 0.2, device)
- self.app_line = self.init_one_svd(self.app_n_comp, self.gridSize, 0.2, device)
+ self.density_line = self.init_one_svd(self.density_n_comp[0], self.gridSize, 0.2, device)
+ self.app_line = self.init_one_svd(self.app_n_comp[0], self.gridSize, 0.2, device)
  self.basis_mat = torch.nn.Linear(self.app_n_comp[0], self.app_dim, bias=False).to(device)
 
 
@@ -321,7 +319,7 @@ def init_one_svd(self, n_component, gridSize, scale, device):
  for i in range(len(self.vecMode)):
  vec_id = self.vecMode[i]
  line_coef.append(
- torch.nn.Parameter(scale * torch.randn((1, n_component[i], gridSize[vec_id], 1))))
+ torch.nn.Parameter(scale * torch.randn((1, n_component, gridSize[vec_id], 1))))
  return torch.nn.ParameterList(line_coef).to(device)
 
 

train.py

@@ -289,8 +289,8 @@ def reconstruction(args):
 if __name__ == '__main__':
 
  torch.set_default_dtype(torch.float32)
- torch.manual_seed(20121202)
- np.random.seed(20121202)
+ torch.manual_seed(20211202)
+ np.random.seed(20211202)
 
  args = config_parser()
  print(args)