fix

config_utils/re_train_autodeeplab.py

@@ -1,79 +1,44 @@
-import  argparse
+import argparse
 
-def obtain_retrain_autodeeplab_args():
-    parser = argparse.ArgumentParser(description="PyTorch DeeplabV3Plus Training")
-    parser.add_argument('--backbone', type=str, default='resnet',
-                        choices=['resnet', 'xception', 'drn', 'mobilenet'],
-                        help='backbone name (default: resnet)')
-    parser.add_argument('--out-stride', type=int, default=16,
-                        help='network output stride (default: 8)')
-    parser.add_argument('--dataset', type=str, default='pascal',
-                        choices=['pascal', 'coco', 'cityscapes'],
-                        help='dataset name (default: pascal)')
-    parser.add_argument('--use-sbd', action='store_true', default=True,
-                        help='whether to use SBD dataset (default: True)')
-    parser.add_argument('--workers', type=int, default=4,
-                        metavar='N', help='dataloader threads')
-    parser.add_argument('--base-size', type=int, default=513,
-                        help='base image size')
-    parser.add_argument('--crop-size', type=int, default=513,
-                        help='crop image size')
-    parser.add_argument('--sync-bn', type=bool, default=None,
-                        help='whether to use sync bn (default: auto)')
-    parser.add_argument('--freeze-bn', type=bool, default=False,
-                        help='whether to freeze bn parameters (default: False)')
-    parser.add_argument('--loss-type', type=str, default='ce',
-                        choices=['ce', 'focal'],
-                        help='loss func type (default: ce)')
-    # training hyper params
-    parser.add_argument('--epochs', type=int, default=None, metavar='N',
-                        help='number of epochs to train (default: auto)')
-    parser.add_argument('--start_epoch', type=int, default=0,
-                        metavar='N', help='start epochs (default:0)')
-    parser.add_argument('--batch-size', type=int, default=None,
-                        metavar='N', help='input batch size for \
-                                training (default: auto)')
-    parser.add_argument('--test-batch-size', type=int, default=None,
-                        metavar='N', help='input batch size for \
-                                testing (default: auto)')
-    parser.add_argument('--use-balanced-weights', action='store_true', default=False,
-                        help='whether to use balanced weights (default: False)')
-    # optimizer params
-    parser.add_argument('--lr', type=float, default=None, metavar='LR',
-                        help='learning rate (default: auto)')
-    parser.add_argument('--lr-scheduler', type=str, default='poly',
-                        choices=['poly', 'step', 'cos'],
-                        help='lr scheduler mode: (default: poly)')
-    parser.add_argument('--momentum', type=float, default=0.9,
-                        metavar='M', help='momentum (default: 0.9)')
-    parser.add_argument('--weight-decay', type=float, default=5e-4,
-                        metavar='M', help='w-decay (default: 5e-4)')
-    parser.add_argument('--nesterov', action='store_true', default=False,
-                        help='whether use nesterov (default: False)')
-    # cuda, seed and logging
-    parser.add_argument('--no-cuda', action='store_true', default=
-    False, help='disables CUDA training')
-    parser.add_argument('--gpu-ids', type=str, default='0',
-                        help='use which gpu to train, must be a \
-                        comma-separated list of integers only (default=0)')
-    parser.add_argument('--seed', type=int, default=1, metavar='S',
-                        help='random seed (default: 1)')
-    # checking point
-    parser.add_argument('--resume', type=str, default=None,
-                        help='put the path to resuming file if needed')
-    parser.add_argument('--checkname', type=str, default=None,
-                        help='set the checkpoint name')
-    # finetuning pre-trained models
-    parser.add_argument('--ft', action='store_true', default=False,
-                        help='finetuning on a different dataset')
-    # evaluation option
-    parser.add_argument('--eval-interval', type=int, default=1,
-                        help='evaluuation interval (default: 1)')
-    parser.add_argument('--no-val', action='store_true', default=False,
-                        help='skip validation during training')
 
+def obtain_retrain_autodeeplab_args():
+    parser = argparse.ArgumentParser(description="PyTorch Autodeeplabv3+ Training")
+    parser.add_argument('--train', action='store_true', default=True, help='training mode')
+    parser.add_argument('--exp', type=str, default='bnlr7e-3', help='name of experiment')
+    parser.add_argument('--gpu', type=str, default='0', help='test time gpu device id')
+    parser.add_argument('--backbone', type=str, default='autodeeplab', help='resnet101')
+    parser.add_argument('--dataset', type=str, default='cityscapes', help='pascal or cityscapes')
+    parser.add_argument('--groups', type=int, default=None, help='num of groups for group normalization')
+    parser.add_argument('--epochs', type=int, default=4000, help='num of training epochs')
+    parser.add_argument('--batch_size', type=int, default=14, help='batch size')
+    parser.add_argument('--base_lr', type=float, default=0.05, help='base learning rate')
+    parser.add_argument('--warmup_start_lr', type=float, default=5e-6, help='warm up learning rate')
+    parser.add_argument('--lr-step', type=float, default=None)
+    parser.add_argument('--warmup-iters', type=int, default=1000)
+    parser.add_argument('--min-lr', type=float, default=None)
+    parser.add_argument('--last_mult', type=float, default=1.0, help='learning rate multiplier for last layers')
+    parser.add_argument('--scratch', action='store_true', default=False, help='train from scratch')
+    parser.add_argument('--freeze_bn', action='store_true', default=False, help='freeze batch normalization parameters')
+    parser.add_argument('--weight_std', action='store_true', default=False, help='weight standardization')
+    parser.add_argument('--beta', action='store_true', default=False, help='resnet101 beta')
+    parser.add_argument('--crop_size', type=int, default=769, help='image crop size')
+    parser.add_argument('--resize', type=int, default=769, help='image crop size')
+    parser.add_argument('--workers', type=int, default=4, help='number of data loading workers')
+    parser.add_argument('--seed', type=int, default=1, metavar='S', help='random seed (default: 1)')
+    parser.add_argument('--resume', type=str, default=None, help='put the path to resuming file if needed')
+    parser.add_argument('--filter_multiplier', type=int, default=20)
+    parser.add_argument('--dist', type=bool, default=False)
+    parser.add_argument('--autodeeplab', type=str, default='train')
+    parser.add_argument('--block_multiplier', type=int, default=5)
     parser.add_argument('--use-ABN', default=True, type=bool, help='whether use ABN')
     parser.add_argument('--affine', default=False, type=bool, help='whether use affine in BN')
-
+    parser.add_argument('--port', default=6000, type=int)
+    parser.add_argument('--max-iteration', default=1000000, type=bool)
+    parser.add_argument('--net_arch', default=None, type=str)
+    parser.add_argument('--cell_arch', default=None, type=str)
+    parser.add_argument('--criterion', default='Ohem', type=str)
+    parser.add_argument('--initial-fm', default=None, type=int)
+    parser.add_argument('--mode', default='poly', type=str, help='how lr decline')
+    parser.add_argument('--local_rank', dest='local_rank', type=int, default=-1, )
     args = parser.parse_args()
-    return args
+    return args

config_utils/re_train_deeplab_v3plus.py

@@ -72,7 +72,6 @@ def obtain_retrain_deeplab_v3plus_args():
                         help='evaluuation interval (default: 1)')
     parser.add_argument('--no-val', action='store_true', default=False,
                         help='skip validation during training')
-
     parser.add_argument('--use-ABN', default=True, type=bool, help='whether use ABN')
     parser.add_argument('--affine', default=False, type=bool, help='whether use affine in BN')
 

config_utils/search_args.py

@@ -97,7 +97,8 @@ def obtain_search_args():
     parser.add_argument('--no-val', action='store_true', default=False,
                         help='skip validation during training')
 
-    parser.add_argument('--use-ABN', default=True, type=bool, help='whether use ABN')
+    parser.add_argument('--use-ABN', default=False, type=bool, help='whether use ABN')
     parser.add_argument('--affine', default=False, type=bool, help='whether use affine in BN')
+    parser.add_argument('--dist', default=False, type=bool, help='whether use dist')
     args = parser.parse_args()
     return args

dataloaders/__init__.py

@@ -1,66 +1,102 @@
 from dataloaders.datasets import cityscapes, kd, coco, combine_dbs, pascal, sbd
 from torch.utils.data import DataLoader
+import torch.utils.data.distributed
+
 
 def make_data_loader(args, **kwargs):
+    if args.dist:
+        print("=> Using Distribued Sampler")
+        if args.dataset == 'cityscapes':
+            if args.autodeeplab == 'search':
+                train_set1, train_set2 = cityscapes.twoTrainSeg(args)
+                num_class = train_set1.NUM_CLASSES
+                sampler1 = torch.utils.data.distributed.DistributedSampler(train_set1)
+                sampler2 = torch.utils.data.distributed.DistributedSampler(train_set2)
+                train_loader1 = DataLoader(train_set1, batch_size=args.batch_size, shuffle=False, sampler=sampler1, **kwargs)
+                train_loader2 = DataLoader(train_set2, batch_size=args.batch_size, shuffle=False, sampler=sampler2, **kwargs)
+
+            elif args.autodeeplab == 'train':
+                train_set = cityscapes.CityscapesSegmentation(args, split='retrain')
+                num_class = train_set.NUM_CLASSES
+                sampler1 = torch.utils.data.distributed.DistributedSampler(train_set)
+                train_loader = DataLoader(train_set, batch_size=args.batch_size, shuffle=False, sampler=sampler1, **kwargs)
+
+            else:
+                raise Exception('autodeeplab param not set properly')
+
+            val_set = cityscapes.CityscapesSegmentation(args, split='val')
+            test_set = cityscapes.CityscapesSegmentation(args, split='test')
+            sampler3 = torch.utils.data.distributed.DistributedSampler(val_set)
+            sampler4 = torch.utils.data.distributed.DistributedSampler(test_set)
+            val_loader = DataLoader(val_set, batch_size=args.batch_size, shuffle=False, sampler=sampler3, **kwargs)
+            test_loader = DataLoader(test_set, batch_size=args.batch_size, shuffle=False, sampler=sampler4, **kwargs)
+
+            if args.autodeeplab == 'search':
+                return train_loader1, train_loader2, val_loader, test_loader, num_class
+            elif args.autodeeplab == 'train':
+                return train_loader, num_class
+        else:
+            raise NotImplementedError
+
+    else:
+        if args.dataset == 'pascal':
+            train_set = pascal.VOCSegmentation(args, split='train')
+            val_set = pascal.VOCSegmentation(args, split='val')
+            if args.use_sbd:
+                sbd_train = sbd.SBDSegmentation(args, split=['train', 'val'])
+                train_set = combine_dbs.CombineDBs([train_set, sbd_train], excluded=[val_set])
 
-    if args.dataset == 'pascal':
-        train_set = pascal.VOCSegmentation(args, split='train')
-        val_set = pascal.VOCSegmentation(args, split='val')
-        if args.use_sbd:
-            sbd_train = sbd.SBDSegmentation(args, split=['train', 'val'])
-            train_set = combine_dbs.CombineDBs([train_set, sbd_train], excluded=[val_set])
-
-        num_class = train_set.NUM_CLASSES
-        train_loader = DataLoader(train_set, batch_size=args.batch_size, shuffle=True, **kwargs)
-        val_loader = DataLoader(val_set, batch_size=args.batch_size, shuffle=False, **kwargs)
-        test_loader = None
-
-        return train_loader, train_loader, val_loader, test_loader, num_class
-
-    elif args.dataset == 'cityscapes':
-        if args.autodeeplab == 'search':
-            train_set1, train_set2 = cityscapes.twoTrainSeg(args)
-            num_class = train_set1.NUM_CLASSES
-            train_loader1 = DataLoader(train_set1, batch_size=args.batch_size, shuffle=True, **kwargs)
-            train_loader2 = DataLoader(train_set2, batch_size=args.batch_size, shuffle=True, **kwargs)
-        elif args.autodeeplab == 'train':
-            train_set = cityscapes.CityscapesSegmentation(args, split='train')
             num_class = train_set.NUM_CLASSES
             train_loader = DataLoader(train_set, batch_size=args.batch_size, shuffle=True, **kwargs)
-        else:
-            raise Exception('autodeeplab param not set properly')
+            val_loader = DataLoader(val_set, batch_size=args.batch_size, shuffle=False, **kwargs)
+            test_loader = None
 
-        val_set = cityscapes.CityscapesSegmentation(args, split='val')
-        test_set = cityscapes.CityscapesSegmentation(args, split='test')
-        val_loader = DataLoader(val_set, batch_size=args.batch_size, shuffle=False, **kwargs)
-        test_loader = DataLoader(test_set, batch_size=args.batch_size, shuffle=False, **kwargs)
+            return train_loader, train_loader, val_loader, test_loader, num_class
 
-        if args.autodeeplab == 'search':
-            return train_loader1, train_loader2, val_loader, test_loader, num_class
-        elif args.autodeeplab == 'train':
-            return train_loader, val_loader, test_loader, num_class
-
-
-    elif args.dataset == 'coco':
-        train_set = coco.COCOSegmentation(args, split='train')
-        val_set = coco.COCOSegmentation(args, split='val')
-        num_class = train_set.NUM_CLASSES
-        train_loader = DataLoader(train_set, batch_size=args.batch_size, shuffle=True, **kwargs)
-        val_loader = DataLoader(val_set, batch_size=args.batch_size, shuffle=False, **kwargs)
-        test_loader = None
-        return train_loader, train_loader, val_loader, test_loader, num_class
-
-    elif args.dataset == 'kd':
-        train_set = kd.CityscapesSegmentation(args, split='train')
-        val_set = kd.CityscapesSegmentation(args, split='val')
-        test_set = kd.CityscapesSegmentation(args, split='test')
-        num_class = train_set.NUM_CLASSES
-        train_loader1 = DataLoader(train_set, batch_size=args.batch_size, shuffle=True, **kwargs)
-        train_loader2 = DataLoader(train_set, batch_size=args.batch_size, shuffle=True, **kwargs)
-        val_loader = DataLoader(val_set, batch_size=args.batch_size, shuffle=False, **kwargs)
-        test_loader = DataLoader(test_set, batch_size=args.batch_size, shuffle=False, **kwargs)
-
-        return train_loader1, train_loader2, val_loader, test_loader, num_class
-    else:
-        raise NotImplementedError
+        elif args.dataset == 'cityscapes':
+            if args.autodeeplab == 'search':
+                train_set1, train_set2 = cityscapes.twoTrainSeg(args)
+                num_class = train_set1.NUM_CLASSES
+                train_loader1 = DataLoader(train_set1, batch_size=args.batch_size, shuffle=True, **kwargs)
+                train_loader2 = DataLoader(train_set2, batch_size=args.batch_size, shuffle=True, **kwargs)
+            elif args.autodeeplab == 'train':
+                train_set = cityscapes.CityscapesSegmentation(args, split='retrain')
+                num_class = train_set.NUM_CLASSES
+                train_loader = DataLoader(train_set, batch_size=args.batch_size, shuffle=True, **kwargs)
+            else:
+                raise Exception('autodeeplab param not set properly')
 
+            val_set = cityscapes.CityscapesSegmentation(args, split='val')
+            test_set = cityscapes.CityscapesSegmentation(args, split='test')
+            val_loader = DataLoader(val_set, batch_size=args.batch_size, shuffle=False, **kwargs)
+            test_loader = DataLoader(test_set, batch_size=args.batch_size, shuffle=False, **kwargs)
+
+            if args.autodeeplab == 'search':
+                return train_loader1, train_loader2, val_loader, test_loader, num_class
+            elif args.autodeeplab == 'train':
+                return train_loader, num_class
+
+
+
+        elif args.dataset == 'coco':
+            train_set = coco.COCOSegmentation(args, split='train')
+            val_set = coco.COCOSegmentation(args, split='val')
+            num_class = train_set.NUM_CLASSES
+            train_loader = DataLoader(train_set, batch_size=args.batch_size, shuffle=True, **kwargs)
+            val_loader = DataLoader(val_set, batch_size=args.batch_size, shuffle=False, **kwargs)
+            test_loader = None
+            return train_loader, train_loader, val_loader, test_loader, num_class
+
+        elif args.dataset == 'kd':
+            train_set = kd.CityscapesSegmentation(args, split='train')
+            val_set = kd.CityscapesSegmentation(args, split='val')
+            test_set = kd.CityscapesSegmentation(args, split='test')
+            num_class = train_set.NUM_CLASSES
+            train_loader1 = DataLoader(train_set, batch_size=args.batch_size, shuffle=True, **kwargs)
+            train_loader2 = DataLoader(train_set, batch_size=args.batch_size, shuffle=True, **kwargs)
+            val_loader = DataLoader(val_set, batch_size=args.batch_size, shuffle=False, **kwargs)
+            test_loader = DataLoader(test_set, batch_size=args.batch_size, shuffle=False, **kwargs)
+
+            return train_loader1, train_loader2, val_loader, test_loader, num_class
+        else:
+            raise NotImplementedError

dataloaders/datasets/cityscapes.py

@@ -1,10 +1,8 @@
 import os
 import numpy as np
-import scipy.misc as m
 from PIL import Image
-from torch.utils import data
 from mypath import Path
-from torchvision import transforms
+from torch.utils import data
 from dataloaders import custom_transforms as tr
 
 

mypath.py

@@ -6,8 +6,7 @@ def db_root_dir(dataset):
         elif dataset == 'sbd':
             return '/path/to/datasets/benchmark_RELEASE/'  # folder that contains dataset/.
         elif dataset == 'cityscapes':
-            # return '/data/deeplearning/cityscapes/'     # foler that contains leftImg8bit/
-            return r'E:\BaiduNetdiskDownload\cityscapes'     # foler that contains leftImg8bit/
+            return '/data/deeplearning/cityscapes/'     # foler that contains leftImg8bit/
         elif dataset == 'kd':
             return '/data/deeplearning/cityscapes/'
         elif dataset == 'coco':