NgeNet renamed with GCNet

README.md

@@ -44,7 +44,7 @@ Paper "Neighborhood-aware Geometric Encoding Network for Point Cloud Registratio
     ```
     # Compile
 
-    cd NgeNet/cpp_wrappers
+    cd GCNet/cpp_wrappers
     sh compile_wrappers.sh
     ```
 
@@ -81,7 +81,7 @@ Unzip it, then we should get the following directories structure:
 # exp_dir: your_saved_path for checkpoints and summary.
 # root: your_data_path for the 3dMatch dataset.
 
-cd NgeNet
+cd GCNet
 python train.py configs/threedmatch.yaml
 
 # note: The code `torch.cuda.empty_cache()` in `train.py` has some impact on the training speed.
@@ -91,7 +91,7 @@ python train.py configs/threedmatch.yaml
 ### 1.3 evaluate and visualize
 
 ```
-cd NgeNet
+cd GCNet
 
 python eval_3dmatch.py --benchmark 3DMatch --data_root your_path/indoor --checkpoint your_path/3dmatch.pth --saved_path work_dirs/3dmatch [--vis] [--no_cuda]
 
@@ -123,14 +123,14 @@ Download odometry kitti [here](http://www.cvlibs.net/datasets/kitti/eval_odometr
 # exp_dir: your_saved_path for checkpoints and summary.
 # root: your_data_path for the Odometry KITTI.
 
-cd NgeNet
+cd GCNet
 python train.py configs/kitti.yaml
 ```
 
 ### 2.3 evaluate and visualize
 
 ```
-cd NgeNet
+cd GCNet
 python eval_kitti.py --data_root your_path/kitti --checkpoint your_path/kitti.pth [--vis] [--no_cuda]
 ```
 

demo.py

@@ -8,15 +8,15 @@
 sys.path.append(os.path.dirname(os.path.abspath(__file__)))
 
 from data import collate_fn
-from models import architectures, NgeNet, vote
+from models import architectures, GCNet, vote
 from utils import decode_config, npy2pcd, pcd2npy, execute_global_registration, \
                   npy2feat, setup_seed, get_blue, get_yellow, voxel_ds, normal, \
                   read_cloud, vis_plys
 
 CUR = os.path.dirname(os.path.abspath(__file__))
 
 
-class NgeNet_pipeline():
+class GCNet_pipeline():
     def __init__(self, ckpt_path, voxel_size, vote_flag, cuda=True):
         self.voxel_size_3dmatch = 0.025
         self.voxel_size = voxel_size
@@ -25,7 +25,7 @@ def __init__(self, ckpt_path, voxel_size, vote_flag, cuda=True):
         self.vote_flag = vote_flag
         config = self.prepare_config()
         self.neighborhood_limits = [38, 36, 35, 38]
-        model = NgeNet(config)
+        model = GCNet(config)
         if self.cuda:
             model = model.cuda()
             model.load_state_dict(torch.load(ckpt_path))
@@ -178,7 +178,7 @@ def pipeline(self, source, target, npts=20000):
     # loading model
     cuda = not args.no_cuda
     vote_flag = not args.no_vote
-    model = NgeNet_pipeline(
+    model = GCNet_pipeline(
         ckpt_path=args.checkpoint, 
         voxel_size=args.voxel_size, 
         vote_flag=vote_flag,

eval_3dmatch.py

@@ -10,7 +10,7 @@
 from tqdm import tqdm
 
 from data import ThreeDMatch, get_dataloader
-from models import architectures, NgeNet, vote
+from models import architectures, GCNet, vote
 from utils import decode_config, npy2pcd, pcd2npy, execute_global_registration, \
                   npy2feat, vis_plys, setup_seed, fmat, to_tensor, get_blue, \
                   get_yellow
@@ -93,7 +93,7 @@ def main(args):
                                                           shuffle=False,
                                                           neighborhood_limits=None)
 
-    model = NgeNet(config)
+    model = GCNet(config)
     use_cuda = not args.no_cuda
     if use_cuda:
         model = model.cuda()

eval_kitti.py

@@ -10,7 +10,7 @@
 from tqdm import tqdm
 
 from data import Kitti, get_dataloader
-from models import architectures, NgeNet, vote
+from models import architectures, GCNet, vote
 from utils import decode_config, npy2pcd, pcd2npy, execute_global_registration, \
     npy2feat, vis_plys, setup_seed, fmat, to_tensor, get_blue, get_yellow
 from metrics import Error_R, Error_t
@@ -39,7 +39,7 @@ def main(args):
                                                           neighborhood_limits=None)
 
     print(neighborhood_limits)
-    model = NgeNet(config)
+    model = GCNet(config)
     use_cuda = not args.no_cuda
     if use_cuda:
         model = model.cuda()

eval_mvp_rg.py

@@ -10,7 +10,7 @@
 from tqdm import tqdm
 
 from data import MVP_RG, get_dataloader
-from models import architectures, NgeNet, vote
+from models import architectures, GCNet, vote
 from utils import decode_config, npy2pcd, pcd2npy, execute_global_registration, \
     npy2feat, vis_plys, setup_seed, fmat, to_tensor, get_blue, get_yellow
 from metrics import Error_R, Error_t, RMSE
@@ -38,7 +38,7 @@ def main(args):
                                                           neighborhood_limits=None)
 
     print(neighborhood_limits)
-    model = NgeNet(config)
+    model = GCNet(config)
     use_cuda = not args.no_cuda
     if use_cuda:
         model = model.cuda()

misc/plot_fmr.py

@@ -125,7 +125,7 @@
 plt.plot(x, dsn_, color='lightseagreen', linewidth=3, label='3DSN', linestyle=linestyle_tuple[4][-1])
 plt.plot(x, ppffold_, color='royalblue', linewidth=3, label='PPF-FoldNet', linestyle=linestyle_tuple[5][-1])
 plt.plot(x, cofinet_, color='tomato', linewidth=3, label='CoFiNet', linestyle='dashdot')
-plt.plot(x, ngenet_, color='magenta', linewidth=3, label='NgeNet', linestyle='solid')
+plt.plot(x, ngenet_, color='magenta', linewidth=3, label='GCNet', linestyle='solid')
 plt.vlines(0.05, 0, 100, colors="gray", linestyles="dashed")
 plt.xlim((0.01, 0.2))
 plt.ylim((0, 100))

models/NgeNet.py → models/GCNet.py

@@ -7,7 +7,7 @@
 from models import InformationInteractive
 
 
-class NgeNet(nn.Module):
+class GCNet(nn.Module):
     def __init__(self, config):
         super().__init__()
         r = config.first_subsampling_dl * config.conv_radius
@@ -218,4 +218,4 @@ def forward(self, inputs):
         batched_feats_l = batched_feats_l / torch.norm(batched_feats_l, dim=1, keepdim=True)
         batched_feats = torch.cat([batched_feats, overlap_scores, saliency_scores], dim=-1)
 
-        return batched_feats, batched_feats_m, batched_feats_l
+        return batched_feats, batched_feats_m, batched_feats_l

models/__init__.py

@@ -1,4 +1,4 @@
 from .architecures import architectures
 from .information_interactive import InformationInteractive
 from .vote import vote
-from .NgeNet import NgeNet
+from .GCNet import GCNet

train.py

@@ -8,7 +8,7 @@
 from torch.utils.tensorboard import SummaryWriter
 
 from data import get_dataset, get_dataloader
-from models import architectures, NgeNet
+from models import architectures, GCNet
 from losses import Loss
 from utils import decode_config, setup_seed
 
@@ -50,7 +50,7 @@ def main():
                                        neighborhood_limits=neighborhood_limits)
 
     print(neighborhood_limits)
-    model = NgeNet(config).cuda()
+    model = GCNet(config).cuda()
     model_loss = Loss(config)
 
     if config.optimizer == 'SGD':