restructure semi-supervised code and remove minor run-time errors

data_loader.py

@@ -1,17 +1,16 @@
 import os
+import pdb
 
 import random
 import torch
 from torchvision import transforms, datasets
 from PIL import ImageFilter
 
-from .util import subset_classes
+from util import subset_classes
 
 
 # Extended version of ImageFolder to return index of image too.
 class ImageFolderEx(datasets.ImageFolder):
-    # def __init__(self, root, sup_split_file, only_sup, *args, **kwargs):
-        # super(ImageFolderEx, self).__init__(root, *args, **kwargs)
     def __init__(self, root, transforms, sup_split_file=None, only_sup=False, corrupt_split_file=None):
         super(ImageFolderEx, self).__init__(root, transforms)
 
@@ -145,15 +144,19 @@ def get_train_loader(opt):
     # Applicable only for semi-supervised setup.
     if 'sup_split_file' in vars(opt).keys():
         # Get dataloader for pseudo-labelling
-        sup_val_dataset = ImageFolderEx(traindir, opt.sup_split_file, True, transforms.Compose(augmentation_weak))
+        sup_val_dataset = ImageFolderEx(
+            root=traindir,
+            sup_split_file=opt.sup_split_file,
+            only_sup=True,
+            transforms=transforms.Compose(augmentation_weak)
+        )
         if opt.dataset == 'imagenet100':
             subset_classes(sup_val_dataset, num_classes=100)
         train_val_loader = torch.utils.data.DataLoader(
             sup_val_dataset,
             batch_size=opt.batch_size, shuffle=False,
             num_workers=opt.num_workers, pin_memory=True,
         )
-
         return train_loader, train_val_loader
     else:
         return train_loader

semi_supervised/run_fullprecision.sh → run_fullprecision.sh

@@ -6,8 +6,9 @@ set -e
 base_dir='./'
 exp_name='exp_semi_sup_fullprec_1'
 dataset_path='path/to/imagenet/dataset'
+#dataset_path='/datasets/imagenet'
 
-CUDA_VISIBLE_DEVICES=0,1,2,3 python train_pseudo_cmsf.py \
+CUDA_VISIBLE_DEVICES=0,1,2,3 python -m semi_supervised.train_pseudo_cmsf \
    --base-dir $base_dir \
    --exp $exp_name\
    --learning_rate 0.05\
@@ -29,10 +30,10 @@ CUDA_VISIBLE_DEVICES=0,1,2,3 python train_pseudo_cmsf.py \
    --mem_bank_size 128000 \
    --sup_mem_bank_size 128000 \
    --save_freq 10\
-   --sup-split-file 'imagenet_subsets/1p_10p/subsets/10percent.txt' \
+   --sup-split-file 'semi_supervised/imagenet_subsets/1p_10p/subsets/10percent.txt' \
    $dataset_path
 
-exp="$base_dir/semi_sup_cmsf/exp/$exp_name"
+exp="$base_dir/exp/semi_sup_cmsf/$exp_name"
 ep=200
 
 CUDA_VISIBLE_DEVICES=0,1 python eval_linear.py\

semi_supervised/pseudo_cmsf.py

@@ -4,10 +4,10 @@
 import torch.nn as nn
 import torch.nn.functional as F
 
-import .models.resnet as resnet
-from .models.mlp_arch import get_mlp, get_mlp_3l
+import models.resnet as resnet
+from models.mlp_arch import get_mlp, get_mlp_3l
 
-from .util import get_shuffle_ids
+from util import get_shuffle_ids
 
 
 class PseudoCMSF(nn.Module):

semi_supervised/pseudo_label_train.py

@@ -3,7 +3,8 @@
 import torch
 import torch.nn.functional as F
 
-from util import AverageMeterV2, ProgressMeter, accuracy
+from util import AverageMeterV2, accuracy
+from tools import ProgressMeter
 
 
 def train_pseudo_lbl(pseudo_cmsf, train_val_loader, backbone, model, logger, opt):

semi_supervised/train_pseudo_cmsf.py

@@ -13,11 +13,11 @@
 import torch
 import torch.backends.cudnn as cudnn
 
-from .data_loader import get_train_loader
-from pseudo_cmsf import PseudoCMSF
-from pseudo_label_train import train_pseudo_lbl
-from .util import adjust_learning_rate, AverageMeter
-from .tools import get_logger
+from data_loader import get_train_loader
+from semi_supervised.pseudo_cmsf import PseudoCMSF
+from semi_supervised.pseudo_label_train import train_pseudo_lbl
+from util import adjust_learning_rate, AverageMeter
+from tools import get_logger
 
 
 def parse_option():

util.py

@@ -44,6 +44,30 @@ def update(self, val, n=1):
         self.avg = self.sum / self.count
 
 
+class AverageMeterV2(object):
+    """Computes and stores the average and current value"""
+    def __init__(self, name, fmt=':f'):
+        self.name = name
+        self.fmt = fmt
+        self.reset()
+
+    def reset(self):
+        self.val = 0
+        self.avg = 0
+        self.sum = 0
+        self.count = 0
+
+    def update(self, val, n=1):
+        self.val = val
+        self.sum += val * n
+        self.count += n
+        self.avg = self.sum / self.count
+
+    def __str__(self):
+        fmtstr = '{name} {val' + self.fmt + '} ({avg' + self.fmt + '})'
+        return fmtstr.format(**self.__dict__)
+
+
 def accuracy(output, target, topk=(1,)):
     """Computes the accuracy over the k top predictions for the specified values of k"""
     with torch.no_grad():
@@ -56,7 +80,7 @@ def accuracy(output, target, topk=(1,)):
 
         res = []
         for k in topk:
-            correct_k = correct[:k].view(-1).float().sum(0, keepdim=True)
+            correct_k = correct[:k].reshape(-1).float().sum(0, keepdim=True)
             res.append(correct_k.mul_(100.0 / batch_size))
         return res