Add Tunnel dataset in pipeline

README.md

@@ -2,6 +2,30 @@
 
 This project has been developed during my internship at LISSI. The objective was to create a pipeline for semantic segmentation with python and pytorch.
 
+
+## :warning: Prevent batch size problem for training :warning:
+
+When you choose the batch size, you need to be careful about the value :
+* `args.batch_size` != 1
+*  dataset size modulo `agrs.batch_size` !=1
+
+
+## Path format for models saves
+
+The train script create a new directory for each execution using the date.
+* 📁 pipeline/
+    * 📁 save/
+        * 📁 model/
+            * 📁 yyyy-mm-dd-id/
+                * best_weights.pth.tar
+                * checkpoint.pth.tar
+                * learning_curves.png
+            * 📁 yyyy-mm-dd-id/
+                * ...
+        * 📁 model/
+            * 📁 yyyy-mm-dd-id/
+                * ...
+
 ## Pipeline architecture
 
 * 📁 pipeline/
@@ -13,6 +37,7 @@ This project has been developed during my internship at LISSI. The objective was
         * 📄 cityscapes.py
         * 📄 minicity.py
         * 📄 learner.py
+        * 📄 tunnel.py
         * 📄 utils.py
     * 📁 models/
         * 📁 configs/
@@ -24,3 +49,4 @@ This project has been developed during my internship at LISSI. The objective was
         * 📄 DenseASPP.py
         * 📄 MobileNetDenseASPP.py
         * 📄 UNet.py
+        * 📄 utils.py

helpers/ARGS.py

@@ -4,9 +4,9 @@
 #  Created by Clément Malonda on 13/07/2021.
 
 class ARGS:
-  def __init__(self):
-    self.dataset_path = "minicity"
-    self.batch_size = 2
+  def __init__(self, model, dataset_path, num_classes, batch_size=8, epochs=80):
+    self.dataset_path = dataset_path
+    self.batch_size = batch_size
     self.pin_memory = True
     self.num_workers = 8
     self.colorjitter_factor = 0.3
@@ -16,9 +16,10 @@ def __init__(self):
     self.dataset_mean = [0.485, 0.456, 0.406]
     self.dataset_std = [0.229, 0.224, 0.225]
     self.loss = "ce"
-    self.model = ""
-    self.copyblob = True
-    self.cutmix = True
-    self.epochs = 4
+    self.model = model
+    self.copyblob = False
+    self.cutmix = False
+    self.epochs = epochs
     self.save_path = ""
     self.is_pytorch_model = True
+    self.num_classes = 30

learning/learner.py

@@ -145,23 +145,6 @@ def validate_epoch(dataloader, model, criterion, epoch, classLabels, validClasse
             # Calculate IoU scores of current batch
             iou.evaluateBatch(preds, labels)
 
-            # Save visualizations of first batch
-            if epoch_step == 0 and maskColors is not None:
-                for i in range(inputs.size(0)):
-                    filename = os.path.splitext(os.path.basename(filepath[i]))[0]
-                    # Only save inputs and labels once
-                    if epoch == 0:
-                        img = visim(inputs[i,:,:,:], args)
-                        label = vislbl(labels[i,:,:], maskColors)
-                        if len(img.shape) == 3:
-                            cv2.imwrite(folder + '/images/{}.png'.format(filename),img[:,:,::-1])
-                        else:
-                            cv2.imwrite(folder + '/images/{}.png'.format(filename),img)
-                        cv2.imwrite(folder + '/images/{}_gt.png'.format(filename),label[:,:,::-1])
-                    # Save predictions
-                    pred = vislbl(preds[i,:,:], maskColors)
-                    cv2.imwrite(folder + '/images/{}_epoch_{}.png'.format(filename,epoch),pred[:,:,::-1])
-
             # measure elapsed time
             batch_time.update(time.time() - end)
             end = time.time()

learning/tunnel.py

@@ -0,0 +1,80 @@
+#
+#  learning/tunnel.py
+#
+#  Created by Clément Malonda on 28/07/2021.
+
+import os
+import json
+import numpy as np
+from PIL import Image
+from PIL import ImageDraw
+import matplotlib.pyplot as plt
+
+classToVal = {"BAU": 1, "Parking" : 2, "SW": 3, "Road": 4}
+
+class Tunnel():
+
+    voidClass = 5
+
+    id2trainid = np.array([0,1,2,3,4], dtype="uint8")
+
+    mask_colors = np.array([[0,0,0], [81,0,81], [230,150,140], [244,35,232], [128,64,128]])
+
+    validClasses = [0,1,2,3,4]
+
+    classLabels = ["Background", "BAU", "Parking", "SW", "Road"]
+
+    def __init__(self, root, split="train", transform=None, target_transform=None, transforms=None):
+        self.root = root
+        self.transform = transform
+        self.target_transform = target_transform
+        self.transforms = transforms
+
+        self.images_dir = os.path.join(self.root, "images")
+        self.target_dir = os.path.join(self.root, "targets")
+        self.split = split
+        self.images = list()
+        self.targets = dict()
+
+        assert split in ["train","val","test"], "Unknown value {} for argument split.".format(split)
+
+        for e in os.listdir(self.target_dir):
+            if e[-5:] == ".json":
+                file = open(os.path.join(self.target_dir, e))
+                targets_data = json.load(file)
+                for target_data in targets_data:
+                    if target_data["name"] in os.listdir(self.images_dir):
+                        self.images.append(os.path.join(self.images_dir, target_data["name"]))
+                        poly = list()
+                        for label in target_data["labels"]["polygonLabels"]:
+                            x = np.array(label["allX"], dtype="int32")
+                            y = np.array(label["allY"], dtype="int32")
+                            points = list(zip(x, y))
+                            poly.append((points, label["labelValue"]))
+                        self.targets.update({os.path.join(self.images_dir, target_data["name"]) :  poly})
+
+                file.close()
+
+
+    def __getitem__(self, index):
+        filepath = self.images[index]
+        image = Image.open(filepath).convert("RGB")
+        # target = np.zeros((image.size[1], image.size[0]))
+        target = Image.new("RGB", (image.size[0], image.size[1]))
+        draw = ImageDraw.Draw(target)
+        for e in self.targets[filepath]:
+            draw.polygon(e[0], fill=classToVal[e[1]])
+        image = np.array(image)
+        image = image.transpose(2, 0, 1)
+        target = np.array(target)[:, :, 0]
+        return image, target, filepath
+
+    def __len__(self):
+        return len(self.images)
+
+if __name__ == "__main__":
+    dataset = Tunnel
+    t = dataset("/Users/ClementMalonda/Desktop/tunnel")
+    image, target, filepath = t[0]
+    plt.imshow(target)
+    plt.show()

learning/utils.py

@@ -102,13 +102,13 @@ def train_trans(image, mask):
     dataloaders = {}
     dataloaders['train'] = torch.utils.data.DataLoader(trainset,
                batch_size=args.batch_size, shuffle=True,
-               pin_memory=args.pin_memory, num_workers=args.num_workers)
+               pin_memory=args.pin_memory, num_workers=args.num_workers, drop_last=True)
     dataloaders['val'] = torch.utils.data.DataLoader(valset,
                batch_size=args.batch_size, shuffle=False,
-               pin_memory=args.pin_memory, num_workers=args.num_workers)
+               pin_memory=args.pin_memory, num_workers=args.num_workers, drop_last=True)
     dataloaders['test'] = torch.utils.data.DataLoader(testset,
                batch_size=args.batch_size, shuffle=False,
-               pin_memory=args.pin_memory, num_workers=args.num_workers)
+               pin_memory=args.pin_memory, num_workers=args.num_workers, drop_last=True)
 
     return dataloaders
 

models/utils.py

@@ -0,0 +1,87 @@
+#
+#  models/utils.py
+#
+#  Created by Clément Malonda on 30/07/2021.
+
+import os
+import time
+from datetime import date
+
+def get_model(args):
+    if args.model == "DeepLabV3_Resnet50":
+        from torchvision.models.segmentation import deeplabv3_resnet50
+        model = deeplabv3_resnet50(num_classes = args.num_classes)
+        args.is_pytorch_model = True
+
+    elif args.model == "DeepLabV3_Resnet101":
+        from torchvision.models.segmentation import deeplabv3_resnet101
+        model = deeplabv3_resnet101(num_classes = args.num_classes)
+        args.is_pytorch_model = True
+
+    elif args.model == "FCN_Resnet50":
+        from torchvision.models.segmentation import fcn_resnet50
+        model = fcn_resnet50(num_classes = args.num_classes)
+        args.is_pytorch_model = True
+
+    elif args.model == "FCN_Resnet101":
+        from torchvision.models.segmentation import fcn_resnet101
+        model = fcn_resnet101(num_classes = args.num_classes)
+        args.is_pytorch_model = True
+
+    elif args.model == "UNet":
+        from models.UNet import UNet
+        model = UNet(num_classes=args.num_classes)
+        args.is_pytorch_model = False
+
+    elif args.model == "DenseASPP121":
+        from models.DenseASPP import DenseASPP
+        from models.configs.DenseASPP121 import Model_CFG
+        model = DenseASPP(Model_CFG, num_classes=args.num_classes)
+        args.is_pytorch_model = False
+
+    elif args.model == "DenseASPP161":
+        from models.DenseASPP import DenseASPP
+        from models.configs.DenseASPP161 import Model_CFG
+        model = DenseASPP(Model_CFG, num_classes=args.num_classes)
+        args.is_pytorch_model = False
+
+    elif args.model == "DenseASPP169":
+        from models.DenseASPP import DenseASPP
+        from models.configs.DenseASPP169 import Model_CFG
+        model = DenseASPP(Model_CFG, num_classes=args.num_classes)
+        args.is_pytorch_model = False
+
+    elif args.model == "DenseASPP201":
+        from models.DenseASPP import DenseASPP
+        from models.configs.DenseASPP201 import Model_CFG
+        model = DenseASPP(Model_CFG, num_classes=args.num_classes)
+        args.is_pytorch_model = False
+
+    elif args.model == "MobileNetDenseASPP":
+        from models.DenseASPP import DenseASPP
+        from models.configs.MobileNetDenseASPP import Model_CFG
+        model = DenseASPP(Model_CFG, num_classes=args.num_classes)
+        args.is_pytorch_model = False
+
+    else:
+        print("Model {} is not supported".format(args.model))
+        print("Models can only be : \
+                \n\t- DeepLabV3_Resnet50 (pytorch)\
+                \n\t- DeepLabV3_Resnet101 (pytorch)\
+                \n\t- FCN_Resnet50 (pytorch)\
+                \n\t- FCN_Resnet101 (pytorch)\
+                \n\t- UNet (external)\
+                \n\t- DenseASPP121 (external)\
+                \n\t- DenseASPP161 (external)\
+                \n\t- DenseASPP169 (external)\
+                \n\t- DenseASPP201 (external)\
+                \n\t- MobileNetDenseASPP (external)")
+        exit(1)
+
+    dir = date.today().isoformat() + "-" +str(int(time.time()))
+    args.save_path = os.path.join("save", args.model, dir)
+
+    return model, args
+
+def get_optimizer(args):
+    pass