print('==========', datetime.now() - start_time)
if __name__ == "__main__":
Mean = [123.68, 116.78, 103.94]
# Mean = [102.9801, 115.9465, 122.7717][::-1]
config = Config(
False,
Mean,
None,
lr=0.00125,
weight_decay=0.0001,
num_cls=80,
img_max=1333,
img_min=800,
anchor_scales=[[32], [64], [128], [256], [512]],
anchor_ratios=[[0.5, 1, 2], [0.5, 1, 2], [0.5, 1, 2], [0.5, 1, 2],
[0.5, 1, 2]],
fast_n_sample=512,
roi_min_size=[4, 8, 16, 32, 64],
roi_train_pre_nms=12000,
roi_train_post_nms=2000,
roi_test_pre_nms=6000,
roi_test_post_nms=1000,
)
model = Mask_Rcnn
model_file = '/home/zhai/PycharmProjects/Demo35/cascade_rcnn/train_M_GPU_single_node/models/Mask_Rcnn_cascade_4x_360000_1_0.pth'
test(model, config, model_file)
path = '/home/zhai/PycharmProjects/Demo35/cascade_rcnn/data_preprocess/'
Bboxes = [path + 'coco_bboxes_2017.pkl']
img_paths = [path + 'coco_imgpaths_2017.pkl']
masks = [path + 'coco_mask_2017.pkl']
files = [img_paths, Bboxes, masks]
config = Config(True,
Mean,
files,
num_cls=80,
lr=0.00125,
weight_decay=0.0001,
batch_size_per_GPU=2,
gpus=2,
img_max=1333,
img_min=800,
anchor_scales=[[32], [64], [128], [256], [512]],
anchor_ratios=[[0.5, 1, 2], [0.5, 1, 2], [0.5, 1, 2],
[0.5, 1, 2], [0.5, 1, 2]],
fast_n_sample=512,
bias_lr_factor=2,
roi_min_size=[4, 8, 16, 32, 64],
roi_train_pre_nms=12000,
roi_train_post_nms=2000,
roi_test_pre_nms=6000,
roi_test_post_nms=1000)
num_gpus = int(
os.environ["WORLD_SIZE"]) if "WORLD_SIZE" in os.environ else 1
assert num_gpus == config.gpus
print('--------GPUs--------', num_gpus)
step = 0
break
if flag:
break
torch.save(model.state_dict(), './models/vgg16_cascade_final_1.pth')
if __name__ == "__main__":
Mean = [123.68, 116.78, 103.94]
path = '/home/zhai/PycharmProjects/Demo35/cascade_rcnn/data_preprocess/'
Bboxes = [path + 'Bboxes_07.pkl', path + 'Bboxes_12.pkl']
img_paths = [path + 'img_paths_07.pkl', path + 'img_paths_12.pkl']
files = [img_paths, Bboxes]
config = Config(True,
Mean,
files,
lr=0.001,
weight_decay=0.0005,
batch_size_per_GPU=1,
img_max=1000,
img_min=600,
bias_lr_factor=2,
roi_min_size=16)
step = 0
model = Faster_Rcnn
x = 1
pre_model_file = '/home/zhai/PycharmProjects/Demo35/py_Faster_tool/pre_model/vgg16_cf.pth'
model_file = None
train(model, config, step, x, pre_model_file, model_file=model_file)
cv2.imshow('a', im)
cv2.waitKey(2000)
return im
if __name__ == "__main__":
from datetime import datetime
from cascade_rcnn.tool.config import Config
Mean = [123.68, 116.78, 103.94]
path = '/home/zhai/PycharmProjects/Demo35/pytorch_Faster_tool/data_preprocess/'
Bboxes = [path + 'coco_bboxes_2017.pkl']
img_paths = [path + 'coco_imgpaths_2017']
masks = [path + 'coco_mask_2017.pkl']
files = [img_paths, Bboxes, masks]
config = Config(True, Mean, files, lr=0.00125, img_max=1000, img_min=600)
dataset = Read_Data(config)
dataloader = DataLoader(dataset,
batch_size=1,
num_workers=16,
collate_fn=lambda x: x)
dataloader = DataLoader(dataset,
batch_size=2,
collate_fn=func,
shuffle=True,
drop_last=True,
pin_memory=False,
num_workers=1)
c = 0
for i in range(2):
Res = {}
start_time = datetime.now()
for name in names[:m]:
i += 1
print(datetime.now(), i)
im_file = test_dir + name + '.jpg'
img = cv2.imread(im_file)
img = torch.tensor(img)
res = model(img)
res = res.cpu()
res = res.detach().numpy()
Res[name] = res
print('==========', datetime.now() - start_time)
joblib.dump(Res, 'Faster_vgg16_cascade_1.pkl')
GT = joblib.load('../mAP/voc_GT.pkl')
AP = mAP(Res, GT, 20, iou_thresh=0.5, use_07_metric=True, e=0.01)
print(AP)
AP = AP.mean()
print(AP)
if __name__ == "__main__":
Mean = [123.68, 116.78, 103.94]
model = Faster_Rcnn
config = Config(False, Mean, None, img_max=1000, img_min=600, roi_min_size=16)
model_file = '/home/zhai/PycharmProjects/Demo35/cascade_rcnn/train_one_GPU/models/vgg16_cascade_90000_1.pth'
test(model, config, model_file)
if __name__ == "__main__":
Mean = [123.68, 116.78, 103.94]
path = '/home/zhai/PycharmProjects/Demo35/cascade_rcnn/data_preprocess/'
Bboxes = [path + 'Bboxes_07.pkl', path + 'Bboxes_12.pkl']
img_paths = [path + 'img_paths_07.pkl', path + 'img_paths_12.pkl']
files = [img_paths, Bboxes]
config = Config(True,
Mean,
files,
lr=0.001,
weight_decay=0.0005,
batch_size_per_GPU=1,
img_max=1000,
img_min=600,
roi_min_size=16,
roi_train_pre_nms=12000,
roi_train_post_nms=2000,
roi_test_pre_nms=6000,
roi_test_post_nms=300,
bias_lr_factor=2)
step = 0
model = Faster_Rcnn
x = 1
pre_model_file = '/home/zhai/PycharmProjects/Demo35/py_Faster_tool/pre_model/vgg16_cf.pth'
model_file = ''
train(model, config, step, x, pre_model_file, model_file=model_file)