Merge pull request CSAILVision#188 from CSAILVision/hang

fix compatibility with torch1.1 and scipy1.3

Author: hangzhaomit

dataset.py

@@ -1,13 +1,29 @@
 import os
 import json
 import torch
+import lib.utils.data as torchdata
 import cv2
 from torchvision import transforms
 import numpy as np
-from scipy.misc import imresize
+import PIL
 
 
-class BaseDataset(torch.utils.data.Dataset):
+def imresize(im, size, interp='bilinear'):
+    if interp == 'nearest':
+        resample = PIL.Image.NEAREST
+    elif interp == 'bilinear':
+        resample = PIL.Image.BILINEAR
+    elif interp == 'bicubic':
+        resample = PIL.Image.BICUBIC
+    else:
+        raise Exception('resample method undefined!')
+
+    return np.array(
+        PIL.Image.fromarray(im).resize((size[1], size[0]), resample)
+    )
+
+
+class BaseDataset(torchdata.Dataset):
     def __init__(self, odgt, opt, **kwargs):
         # parse options
         self.imgSizes = opt.imgSizes

eval.py

@@ -15,6 +15,7 @@
 from utils import AverageMeter, colorEncode, accuracy, intersectionAndUnion, setup_logger
 from lib.nn import user_scattered_collate, async_copy_to
 from lib.utils import as_numpy
+import lib.utils.data as torchdata
 import cv2
 from tqdm import tqdm
 
@@ -132,7 +133,7 @@ def main(cfg, gpu):
         cfg.DATASET.root_dataset,
         cfg.DATASET.list_val,
         cfg.DATASET)
-    loader_val = torch.utils.data.DataLoader(
+    loader_val = torchdata.DataLoader(
         dataset_val,
         batch_size=cfg.VAL.batch_size,
         shuffle=False,

eval_multipro.py

@@ -16,6 +16,7 @@
 from utils import AverageMeter, colorEncode, accuracy, intersectionAndUnion, parse_devices, setup_logger
 from lib.nn import user_scattered_collate, async_copy_to
 from lib.utils import as_numpy
+import lib.utils.data as torchdata
 import cv2
 from tqdm import tqdm
 
@@ -93,7 +94,7 @@ def worker(cfg, gpu_id, start_idx, end_idx, result_queue):
         cfg.DATASET.list_val,
         cfg.DATASET,
         start_idx=start_idx, end_idx=end_idx)
-    loader_val = torch.utils.data.DataLoader(
+    loader_val = torchdata.DataLoader(
         dataset_val,
         batch_size=cfg.VAL.batch_size,
         shuffle=False,

lib/utils/data/dataloader.py

@@ -1,10 +1,13 @@
 import torch
 import torch.multiprocessing as multiprocessing
-from torch._C import _set_worker_signal_handlers, _update_worker_pids, \
+from torch._C import _set_worker_signal_handlers, \
     _remove_worker_pids, _error_if_any_worker_fails
+try:
+    from torch._C import _set_worker_pids
+except:
+    from torch._C import _update_worker_pids as _set_worker_pids
 from .sampler import SequentialSampler, RandomSampler, BatchSampler
 import signal
-import functools
 import collections
 import re
 import sys
@@ -235,7 +238,7 @@ def __init__(self, loader):
                 w.daemon = True  # ensure that the worker exits on process exit
                 w.start()
 
-            _update_worker_pids(id(self), tuple(w.pid for w in self.workers))
+            _set_worker_pids(id(self), tuple(w.pid for w in self.workers))
             _set_SIGCHLD_handler()
             self.worker_pids_set = True
 

test.py

@@ -14,6 +14,7 @@
 from utils import colorEncode, find_recursive, setup_logger
 from lib.nn import user_scattered_collate, async_copy_to
 from lib.utils import as_numpy
+import lib.utils.data as torchdata
 import cv2
 from tqdm import tqdm
 from config import cfg
@@ -115,7 +116,7 @@ def main(cfg, gpu):
     dataset_test = TestDataset(
         cfg.list_test,
         cfg.DATASET)
-    loader_test = torch.utils.data.DataLoader(
+    loader_test = torchdata.DataLoader(
         dataset_test,
         batch_size=cfg.TEST.batch_size,
         shuffle=False,

train.py

@@ -14,6 +14,7 @@
 from models import ModelBuilder, SegmentationModule
 from utils import AverageMeter, parse_devices, setup_logger
 from lib.nn import UserScatteredDataParallel, user_scattered_collate, patch_replication_callback
+import lib.utils.data as torchdata
 
 
 # train one epoch
@@ -168,7 +169,7 @@ def main(cfg, gpus):
         cfg.DATASET,
         batch_per_gpu=cfg.TRAIN.batch_size_per_gpu)
 
-    loader_train = torch.utils.data.DataLoader(
+    loader_train = torchdata.DataLoader(
         dataset_train,
         batch_size=len(gpus),  # we have modified data_parallel
         shuffle=False,  # we do not use this param