code

.gitignore

@@ -0,0 +1,10 @@
+.vscode/
+/models
+.idea/
+loader/__pycache__/
+network/__pycache__/
+loss/__pycache__/
+/log
+/eval
+/test
+__pycache__/

inference/inference_kaist/inference_sc_kaist1_disco_lidar.py

@@ -0,0 +1,96 @@
+from torch.utils.data import Dataset, DataLoader
+import torch
+import torch.nn as nn
+import os
+import sys
+import time
+import numpy as np
+from torch.utils.tensorboard import SummaryWriter
+
+sys.path.append("..")
+
+from loader.data_loader_inference import dataLoader_inference
+from network.radar_coder import radar_coder
+from network.lidar_coder import lidar_coder
+from loss.fft_operation import fft_operation, fftshift2d
+
+import torchvision
+import torchvision.transforms.functional as TF
+from PIL import Image
+import copy
+
+# 2020.12.24
+SC_H = 40
+SC_W = 120
+# SC_H = 64
+# SC_W = 120
+
+# total_num = 8866  # RobotCar
+total_num = 3298 # KAIST01
+# total_num = 3586 # KAIST02
+# total_num = 2212 # DCC01
+# total_num = 3025 # DCC02
+
+# Data
+
+root_radar_path = '/home/yinhuan/Data/radar_lidar_pr/Mulran/Kaist01_80/radar_sc/'
+# root_lidar_path = '/home/yinhuan/Data/radar_lidar_pr/Mulran/Kaist01_80/lidar_sc_density/'
+root_lidar_path = '/home/yinhuan/Data/radar_lidar_pr/Mulran/Kaist01_80/lidar_sc_occupy/'
+save_lidar_fft_path = '/home/yinhuan/Data/radar_lidar_pr/Mulran/Kaist01_80/lidar_fft_disco_ll_quad/'
+
+# model
+lidar_coder_path = '/home/yinhuan/radar_lidar_place_recognition/models/01.27_radar_lidar_pr_disco_lidar/lidar_coder_model_5.pth'
+
+if __name__ == "__main__":
+
+    device = 'cuda:3'
+
+    np.set_printoptions(precision=5)
+
+    print('Build Model')
+
+    # lidar coder init
+    lidar_coder = lidar_coder().to(device)
+    lidar_coder.load_state_dict(torch.load(lidar_coder_path))
+    lidar_coder.eval()
+    for m in lidar_coder.modules():
+        if isinstance(m, nn.BatchNorm2d):
+            m.track_running_stats=False
+
+    # dataloader init and shuffle
+    dataLoader_ = dataLoader_inference(total_num, root_radar_path, root_lidar_path, SC_H, SC_W)
+
+    pdist = nn.PairwiseDistance(p=2)
+
+    # count the num 
+    for iter_ in range(total_num):
+
+            print('----------------------------------------------------' + str(iter_))
+            t0 = time.time()
+
+            radar_data, lidar_data = dataLoader_.get_data\
+                (iter_+1, device) # from 1, not 0
+
+
+            lidar_feature = lidar_coder(radar_data)
+
+            lidar_fft = fft_operation(lidar_feature)
+
+            lidar_fft = fftshift2d(lidar_fft)
+
+            lidar_fft = lidar_fft[:,:,\
+                 (SC_H//2-16):(SC_H//2+16), (SC_W//2-16):(SC_W//2+16)]
+
+            # save txt
+            if not os.path.exists(save_lidar_fft_path):
+                os.makedirs(save_lidar_fft_path)
+
+            lidar_fft_np = lidar_fft.view(32, 32).detach().cpu().numpy()
+            with open(save_lidar_fft_path + str(iter_+1) + '.txt', 'wb') as f:
+                np.savetxt(f, lidar_fft_np, fmt='%5.5f', delimiter=' ', newline='\n')
+
+            # images = torch.cat([radar_batch[0].unsqueeze(0), radar_feature_batch[0].unsqueeze(0),\
+            #      lidar_batch[0].unsqueeze(0), lidar_feature_batch[0].unsqueeze(0)])
+            # writer.add_images(date+'/Images', images, iter_)
+
+    print('Finished')

inference/inference_kaist/inference_sc_kaist1_disco_radar.py

@@ -0,0 +1,98 @@
+from torch.utils.data import Dataset, DataLoader
+import torch
+import torch.nn as nn
+import os
+import sys
+import time
+import numpy as np
+from torch.utils.tensorboard import SummaryWriter
+
+sys.path.append("..")
+
+from loader.data_loader_inference import dataLoader_inference
+from network.radar_coder import radar_coder
+from network.lidar_coder import lidar_coder
+from loss.fft_operation import fft_operation, fftshift2d
+
+import torchvision
+import torchvision.transforms.functional as TF
+from PIL import Image
+import copy
+
+# 2020.12.24
+SC_H = 40
+SC_W = 120
+# SC_H = 64
+# SC_W = 120
+
+# total_num = 8866  # RobotCar
+total_num = 3298 # KAIST01
+# total_num = 3586 # KAIST02
+# total_num = 2212 # DCC01
+# total_num = 3025 # DCC02
+
+# Data
+
+root_radar_path = '/home/yinhuan/Data/radar_lidar_pr/Mulran/Kaist01_80/radar_sc/'
+# root_lidar_path = '/home/yinhuan/Data/radar_lidar_pr/Mulran/Kaist01_80/lidar_sc_density/'
+root_lidar_path = '/home/yinhuan/Data/radar_lidar_pr/Mulran/Kaist01_80/lidar_sc_occupy/'
+save_radar_fft_path = '/home/yinhuan/Data/radar_lidar_pr/Mulran/Kaist01_80/radar_fft_disco_rr_quad/'
+
+# model
+radar_coder_path = '/home/yinhuan/radar_lidar_place_recognition/models/01.27_radar_lidar_pr_disco_radar/radar_coder_model_5.pth'
+
+if __name__ == "__main__":
+
+    device = 'cuda:3'
+
+    np.set_printoptions(precision=5)
+
+    print('Build Model')
+
+    # radar coder init
+    radar_coder = radar_coder().to(device)
+    radar_coder.load_state_dict(torch.load(radar_coder_path))
+    radar_coder.eval()
+    for m in radar_coder.modules():
+        if isinstance(m, nn.BatchNorm2d):
+            m.track_running_stats=False
+
+    # dataloader init and shuffle
+    dataLoader_ = dataLoader_inference(total_num, root_radar_path, root_lidar_path, SC_H, SC_W)
+
+    pdist = nn.PairwiseDistance(p=2)
+
+    # count the num 
+    for iter_ in range(total_num):
+
+            print('----------------------------------------------------' + str(iter_))
+            t0 = time.time()
+
+            radar_data, lidar_data = dataLoader_.get_data\
+                (iter_+1, device) # from 1, not 0
+
+            # radar_translated = radar_transformer(radar_data)
+            # use both lidar_coder ?
+
+            radar_feature = radar_coder(radar_data)
+
+            radar_fft = fft_operation(radar_feature)
+
+            radar_fft = fftshift2d(radar_fft)
+
+            radar_fft = radar_fft[:,:,\
+                 (SC_H//2-16):(SC_H//2+16), (SC_W//2-16):(SC_W//2+16)]
+
+            # save txt
+            if not os.path.exists(save_radar_fft_path):
+                os.makedirs(save_radar_fft_path)
+
+            radar_fft_np = radar_fft.view(32, 32).detach().cpu().numpy()
+            with open(save_radar_fft_path + str(iter_+1) + '.txt', 'wb') as f:
+                np.savetxt(f, radar_fft_np, fmt='%5.5f', delimiter=' ', newline='\n')
+
+            # images = torch.cat([radar_batch[0].unsqueeze(0), radar_feature_batch[0].unsqueeze(0),\
+            #      lidar_batch[0].unsqueeze(0), lidar_feature_batch[0].unsqueeze(0)])
+            # writer.add_images(date+'/Images', images, iter_)
+
+    print('Finished')

inference/inference_kaist/inference_sc_kaist1_joint.py

@@ -0,0 +1,120 @@
+from torch.utils.data import Dataset, DataLoader
+import torch
+import torch.nn as nn
+import os
+import sys
+import time
+import numpy as np
+from torch.utils.tensorboard import SummaryWriter
+
+sys.path.append("..")
+
+from loader.data_loader_inference import dataLoader_inference
+from network.radar_coder import radar_coder
+from loss.fft_operation import fft_operation, fftshift2d
+
+import torchvision
+import torchvision.transforms.functional as TF
+from PIL import Image
+import copy
+
+# 2020.12.24
+SC_H = 40
+SC_W = 120
+# SC_H = 64
+# SC_W = 120
+
+# total_num = 8866  # RobotCar
+total_num = 3298 # KAIST01
+# total_num = 3586 # KAIST02
+# total_num = 2212 # DCC01
+# total_num = 3025 # DCC02
+
+# Data
+
+root_radar_path = '/home/yinhuan/Data/radar_lidar_pr/Mulran/Kaist01_80/radar_sc/'
+# root_lidar_path = '/home/yinhuan/Data/radar_lidar_pr/Mulran/Kaist01_80/lidar_sc_density/'
+root_lidar_path = '/home/yinhuan/Data/radar_lidar_pr/Mulran/Kaist01_80/lidar_sc_occupy/'
+save_radar_fft_path = '/home/yinhuan/Data/radar_lidar_pr/Mulran/Kaist01_80/radar_fft_disco_joint/'
+save_lidar_fft_path = '/home/yinhuan/Data/radar_lidar_pr/Mulran/Kaist01_80/lidar_fft_disco_joint/'
+
+# model
+radar_coder_path = '/home/yinhuan/radar_lidar_place_recognition/models/01.18_radar_lidar_pr_disco_joint/radar_coder_model_5.pth'
+
+if __name__ == "__main__":
+
+    device = 'cuda:3'
+
+    np.set_printoptions(precision=5)
+
+    print('Build Model')
+
+    # # radar transformer init
+    # radar_transformer = radar_transformer().to(device)
+    # radar_transformer.load_state_dict(torch.load(radar_transformer_path))
+    # radar_transformer.eval()
+    # for m in radar_transformer.modules():
+    #     if isinstance(m, nn.BatchNorm2d):
+    #         m.track_running_stats=False
+
+    # radar coder init
+    radar_coder = radar_coder().to(device)
+    radar_coder.load_state_dict(torch.load(radar_coder_path))
+    radar_coder.eval()
+    for m in radar_coder.modules():
+        if isinstance(m, nn.BatchNorm2d):
+            m.track_running_stats=False
+
+    # dataloader init and shuffle
+    dataLoader_ = dataLoader_inference(total_num, root_radar_path, root_lidar_path, SC_H, SC_W)
+
+    pdist = nn.PairwiseDistance(p=2)
+
+    # count the num 
+    for iter_ in range(total_num):
+
+            print('----------------------------------------------------' + str(iter_))
+            t0 = time.time()
+
+            radar_data, lidar_data = dataLoader_.get_data\
+                (iter_+1, device) # from 1, not 0
+
+            # radar_translated = radar_transformer(radar_data)
+
+            radar_feature = radar_coder(radar_data)
+            lidar_feature = radar_coder(lidar_data)
+
+            radar_fft = fft_operation(radar_feature)
+            lidar_fft = fft_operation(lidar_feature)
+
+            radar_fft = fftshift2d(radar_fft)
+            lidar_fft = fftshift2d(lidar_fft)
+
+            radar_fft = radar_fft[:,:,\
+                 (SC_H//2-16):(SC_H//2+16), (SC_W//2-16):(SC_W//2+16)]
+            lidar_fft = lidar_fft[:,:,\
+                 (SC_H//2-16):(SC_H//2+16), (SC_W//2-16):(SC_W//2+16)]
+
+            # save png
+            # radar_translated_img = TF.to_pil_image(radar_translated[0].cpu())
+            # radar_translated_img.save(save_radar_translated_path + str(iter_+1) + '.png')
+
+            # save txt
+            if not os.path.exists(save_radar_fft_path):
+                os.makedirs(save_radar_fft_path)
+            if not os.path.exists(save_lidar_fft_path):
+                os.makedirs(save_lidar_fft_path)
+
+            radar_fft_np = radar_fft.view(32, 32).detach().cpu().numpy()
+            with open(save_radar_fft_path + str(iter_+1) + '.txt', 'wb') as f:
+                np.savetxt(f, radar_fft_np, fmt='%5.5f', delimiter=' ', newline='\n')
+
+            lidar_fft_np = lidar_fft.view(32, 32).detach().cpu().numpy()
+            with open(save_lidar_fft_path + str(iter_+1) + '.txt', 'wb') as f:
+                np.savetxt(f, lidar_fft_np, fmt='%5.5f', delimiter=' ', newline='\n')
+
+            # images = torch.cat([radar_batch[0].unsqueeze(0), radar_feature_batch[0].unsqueeze(0),\
+            #      lidar_batch[0].unsqueeze(0), lidar_feature_batch[0].unsqueeze(0)])
+            # writer.add_images(date+'/Images', images, iter_)
+
+    print('Finished')