for cluster in range(k):
clusters[cluster] = dist(boxes[nearest_clusters == cluster],
axis=0)
last_clusters = nearest_clusters
return clusters
if __name__ == "__main__":
import sys
import torchvision
import os
from datasets.voc import handle_voc_target, VOCDetection
sys.path.append(os.path.abspath(".."))
root_path = os.path.abspath("..")
sets = VOCDetection(os.path.join(root_path, "data"),
image_set="train",
target_transform=handle_voc_target)
boxes = []
for _, target in sets:
boxes.append(target[:, 1:5])
boxes = np.concatenate(boxes, 0)
boxes = handle_boxes(boxes)
out = kmeans(boxes, k=9)
print("Boxes: \n {}".format(out))
ratios = np.around(out[:, 0] / out[:, 1], decimals=3).tolist()
print("Ratios: \n {}".format(sorted(ratios)))
# =====================================
transform_train = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224,
0.225]),
]) # apply to image data only
transform_test = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224,
0.225]),
]) # apply to image data only
trainset = VOCDetection(root='./datasets',
year='2007',
image_set='train',
download=False,
transform=transform_train)
testset = VOCDetection(root='./datasets',
year='2007',
image_set='val',
download=False,
transform=transform_train)
trainloader = torch.utils.data.DataLoader(trainset,
batch_size=batch_size,
shuffle=True,
num_workers=4)
testloader = torch.utils.data.DataLoader(testset,
def init_dataloader_train(opt):
batch_size = opt.batch_size_train
dataset_name = opt.dataset
transform = init_transforms(opt, 'train')
if dataset_name.startswith('voc'):
opt.num_classes = 21
from datasets.voc import VOCDetection
voc_root = os.path.join(opt.data_root_dir, 'VOCdevkit')
# voc 2007
if dataset_name == 'voc07':
image_root = os.path.join(voc_root, 'VOC2007/JPEGImages')
list_file = os.path.join('data/voc/voc07_trainval.txt')
# voc 2012
elif dataset_name == 'voc12':
image_root = os.path.join(voc_root, 'VOC2012/JPEGImages')
list_file = os.path.join('data/voc/voc12_trainval.txt')
# voc 0712
elif dataset_name == 'voc0712':
image_root = os.path.join(voc_root, 'VOC0712/JPEGImages')
list_file = [os.path.join('data/voc/voc07_trainval.txt'),
os.path.join('data/voc/voc12_trainval.txt')]
else:
raise ValueError('Not a valid dataset name')
dataset = VOCDetection(root=image_root,
list_file=list_file,
transform=transform)
loader = dataloader.DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=True,
num_workers=opt.num_workers)
elif dataset_name == 'wider':
opt.num_classes = 2
from datasets.wider import WiderDetection
wider_root = os.path.join(opt.data_root_dir, 'WIDER/WIDER_train/images')
list_file = 'data/wider/wider_train.txt'
dataset = WiderDetection(root=wider_root,
list_file=list_file,
transform=transform)
loader = dataloader.DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=False,
num_workers=opt.num_workers,)
elif dataset_name == 'fddb':
opt.num_classes = 2
from datasets.fddb import FDDBDetection
from preprocess.fddb.preprocess import parse_fddb_annotation
wider_root = os.path.join(opt.data_root_dir, 'FDDB/imgs')
annotation_dir = os.path.join(opt.data_root_dir, 'FDDB/anno')
list_file = os.path.join(opt.project_root, 'data/fddb/fddb_train.txt')
os.system('rm %s' % list_file)
parse_fddb_annotation(wider_root, annotation_dir, 'train', list_file, opt.fold)
dataset = FDDBDetection(root=wider_root,
list_file=list_file,
mode='train',
transform=transform)
loader = dataloader.DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=True,
num_workers=opt.num_workers)
elif dataset_name == 'v_caption_detection':
from datasets.v_caption_detection import V_Caption_Detection
data_root = os.path.join(opt.data_root_dir, 'V.DO/caption/background_result')
list_file = 'data/v_caption_detection/caption_BG_train.txt'
dataset = V_Caption_Detection(root=data_root,
list_file=list_file,
transform=transform)
loader = dataloader.DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=True,
num_workers=opt.num_workers)
elif dataset_name == 'v_caption_patch':
from datasets.v_caption_patch import V_Caption_Patch
data_root = os.path.join(opt.data_root_dir, '')
list_file = 'data/v_caption_patch_hangul/patch_train.txt'
dataset = V_Caption_Patch(root=data_root,
list_file=list_file,
transform=transform)
loader = dataloader.DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=True,
num_workers=opt.num_workers,
pin_memory=True)
elif dataset_name == 'v_caption':
from datasets.v_caption import V_Caption
data_root = os.path.join(opt.data_root_dir, '')
list_file = ['data/v_caption/hangul_patch_train.txt']
dataset = V_Caption(root=data_root,
list_file=list_file,
transform=transform)
loader = dataloader.DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=True,
num_workers=opt.num_workers,
pin_memory=True)
elif dataset_name == 'v_caption_patch_num':
from datasets.v_caption_patch_num import V_Caption_Patch_Num
data_root = os.path.join(opt.data_root_dir, '')
list_file = 'data/v_caption_detection/bgnumber_patch_train.txt'
dataset = V_Caption_Patch_Num(root=data_root,
list_file=list_file,
transform=transform)
loader = dataloader.DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=True,
num_workers=opt.num_workers,
pin_memory=True)
elif dataset_name == 'v_caption_patch_alp':
from datasets.v_caption_patch_alp import V_Caption_Patch_Alp
data_root = os.path.join(opt.data_root_dir, '')
list_file = 'data/v_caption_detection/bgalphabet_patch_train.txt'
dataset = V_Caption_Patch_Alp(root=data_root,
list_file=list_file,
transform=transform)
loader = dataloader.DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=True,
num_workers=opt.num_workers,
pin_memory=True)
elif dataset_name == 'v_caption_patch_alp2':
from datasets.v_caption_patch_alp2 import V_Caption_Patch_Alp2
data_root = os.path.join(opt.data_root_dir, '')
list_file = ['data/v_caption_detection/bgalphabet_patch_train.txt','data/v_caption_detection/bgalphabet_patch_val.txt','data/v_caption_detection/alphabet_patch_train.txt']
dataset = V_Caption_Patch_Alp2(root=data_root,
list_file=list_file,
transform=transform)
loader = dataloader.DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=True,
num_workers=opt.num_workers,
pin_memory=True)
elif dataset_name == 'v_caption_patch_sym':
from datasets.v_caption_patch_sym import V_Caption_Patch_Sym
data_root = os.path.join(opt.data_root_dir, '')
list_file ='data/v_caption_detection/bgsymbol_patch_train.txt'
dataset = V_Caption_Patch_Sym(root=data_root,
list_file=list_file,
transform=transform)
loader = dataloader.DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=True,
num_workers=opt.num_workers,
pin_memory=True)
elif dataset_name == 'v_caption_patch_type':
opt.num_classes = 4
from datasets.v_caption_patch_type import V_Caption_Resnet_Type
data_root = os.path.join(opt.data_root_dir, '')
list_file = ['data/v_caption_detection/bgalphabet_patch_train.txt','data/v_caption_detection/bghangul_patch_train.txt','data/v_caption_detection/bgnumber_patch_train.txt',
'data/v_caption_detection/bgsymbol_patch_train.txt']
dataset = V_Caption_Resnet_Type(root=data_root,
list_file=list_file,
transform=transform)
loader = dataloader.DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=True,
num_workers=opt.num_workers,
pin_memory=True)
else:
raise ValueError('Not a valid dataset')
return loader
orders = orders[iou <= threshold]
boxes = boxes[iou <= threshold,:]
keeps.append(keep)
return keeps
# Define the recall and total dictionary for calculating mAP
#recall = {}
#total = {}
#for i in range(21):
# recall[i] = 0
# total[i] = 0
# Define the test loader: Only use toTensor to keep the original image
testset = VOCDetection(root='./datasets', year='2007', image_set='train', download=False, transform=transforms.ToTensor())
testloader = torch.utils.data.DataLoader(testset, batch_size=1, shuffle=False, num_workers=1)
# Load the model
checkpoint = torch.load('rcnn.pth', map_location=torch.device('cpu'))
model = FasterRCNN()
model.eval() # Turn to eval mode
model.load_state_dict(checkpoint['state_dict'])
print("Start eval...")
for batch_idx, (images, gt_boxes, gt_classes, gt_boxes_, gt_classes_) in enumerate(testloader):
# Use for partial training
if batch_idx <= 147:
continue
# Add the normalization step
def init_dataloader_valid(opt):
batch_size = opt.batch_size_valid
dataset_name = opt.dataset
transform = init_transforms(opt, 'valid')
if dataset_name.startswith('voc'):
opt.num_classes = 21
from datasets.voc import VOCDetection
voc_root = os.path.join(opt.data_root_dir, 'VOCdevkit')
# voc 2007
if dataset_name == 'voc07':
image_root = os.path.join(voc_root, 'VOC2007/JPEGImages')
list_file = os.path.join('data/voc/voc07_test.txt')
# voc 2012
elif dataset_name == 'voc12':
image_root = os.path.join(voc_root, 'VOC2012/JPEGImages')
list_file = os.path.join('data/voc/voc12_test.txt')
# voc 0712
elif dataset_name == 'voc0712':
image_root = os.path.join(voc_root, 'VOC0712/JPEGImages')
list_file = os.path.join('data/voc/voc07_test.txt')
else:
raise ValueError('Not a valid dataset name')
dataset = VOCDetection(root=image_root,
list_file=list_file,
transform=transform)
loader = dataloader.DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=False,
num_workers=opt.num_workers)
elif dataset_name == 'wider':
opt.num_classes = 2
from datasets.wider import WiderDetection
wider_root = os.path.join(opt.data_root_dir, 'WIDER/WIDER_val/images')
list_file = 'data/wider/wider_val.txt'
dataset = WiderDetection(root=wider_root,
list_file=list_file,
transform=transform)
loader = dataloader.DataLoader(
dataset=dataset,
batch_size=batch_size,
shuffle=False,
num_workers=opt.num_workers,
)
elif dataset_name == 'fddb':
opt.num_classes = 2
from datasets.fddb import FDDBDetection
from preprocess.fddb.preprocess import parse_fddb_annotation
wider_root = os.path.join(opt.data_root_dir, 'FDDB/imgs')
annotation_dir = os.path.join(opt.data_root_dir, 'FDDB/anno')
list_file = os.path.join(opt.project_root, 'data/fddb/fddb_val.txt')
os.system('rm %s' % list_file)
parse_fddb_annotation(wider_root, annotation_dir, 'valid', list_file,
opt.fold)
dataset = FDDBDetection(root=wider_root,
list_file=list_file,
mode='valid',
transform=transform)
loader = dataloader.DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=False,
num_workers=opt.num_workers)
elif dataset_name == 'v_caption':
from datasets.v_caption import V_Caption
data_root = os.path.join(opt.data_root_dir, '')
list_file = ['data/v_caption/bghangul_patch_val.txt']
dataset = V_Caption(root=data_root,
list_file=list_file,
transform=transform)
loader = dataloader.DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=False,
num_workers=opt.num_workers,
pin_memory=True)
else:
raise ValueError('Not a valid dataset')
return loader