def run_demo(cfg, ckpt, score_threshold, images_dir, output_dir, dataset_type):
if dataset_type == "voc":
class_names = VOCDataset.class_names
elif dataset_type == 'coco':
class_names = COCODataset.class_names
else:
raise NotImplementedError('Not implemented now.')
device = torch.device(cfg.MODEL.DEVICE)
model = build_detection_model(cfg)
model = model.to(device)
checkpointer = CheckPointer(model, save_dir=cfg.OUTPUT_DIR)
checkpointer.load(ckpt, use_latest=ckpt is None)
weight_file = ckpt if ckpt else checkpointer.get_checkpoint_file()
print('Loaded weights from {}'.format(weight_file))
# dummy_input = torch.randn(1, 3, 300, 300, device='cuda')
# input_names = ["input"]
# output_names = ["output"]
# torch.onnx.export(model, dummy_input, "vgg_ssd300_voc.onnx", verbose=True, input_names=input_names, output_names=output_names)
image_paths = glob.glob(os.path.join(images_dir, '*.jpg'))
mkdir(output_dir)
cpu_device = torch.device("cpu")
transforms = build_transforms(cfg, is_train=False)
model.eval()
for i, image_path in enumerate(image_paths):
start = time.time()
image_name = os.path.basename(image_path)
image = np.array(Image.open(image_path).convert("RGB"))
height, width = image.shape[:2]
images = transforms(image)[0].unsqueeze(0)
load_time = time.time() - start
start = time.time()
result = model(images.to(device))[0]
inference_time = time.time() - start
result = result.resize((width, height)).to(cpu_device).numpy()
boxes, labels, scores = result['boxes'], result['labels'], result[
'scores']
indices = scores > score_threshold
boxes = boxes[indices]
labels = labels[indices]
scores = scores[indices]
meters = ' | '.join([
'objects {:02d}'.format(len(boxes)),
'load {:03d}ms'.format(round(load_time * 1000)),
'inference {:03d}ms'.format(round(inference_time * 1000)),
'FPS {}'.format(round(1.0 / inference_time))
])
print('({:04d}/{:04d}) {}: {}'.format(i + 1, len(image_paths),
image_name, meters))
drawn_image = draw_boxes(image, boxes, labels, scores,
class_names).astype(np.uint8)
Image.fromarray(drawn_image).save(os.path.join(output_dir, image_name))
def do_run(cfg, model, distributed, **kwargs):
if isinstance(model, torch.nn.parallel.DistributedDataParallel):
model = model.module
model.eval()
device = torch.device(cfg.MODEL.DEVICE)
data_loaders_val = make_data_loader(cfg,
is_train=False,
distributed=distributed)
for dataset_name, data_loader in zip(cfg.DATASETS.TEST, data_loaders_val):
output_folder = os.path.join(cfg.OUTPUT_DIR, "inference", dataset_name)
if not os.path.exists(output_folder):
mkdir(output_folder)
run(model, data_loader, dataset_name, device, output_folder, **kwargs)
def do_evaluation(cfg, model, distributed, **kwargs):
if isinstance(model, torch.nn.parallel.DistributedDataParallel):
model = model.module
model.eval()
device = torch.device(cfg.MODEL.DEVICE)
data_loaders_val = make_data_loader(cfg,
is_train=False,
distributed=distributed)
eval_results = []
timer = Timer()
timer.tic()
for dataset_name, data_loader in zip(cfg.DATASETS.TEST, data_loaders_val):
output_folder = os.path.join(cfg.OUTPUT_DIR, "inference", dataset_name)
if not os.path.exists(output_folder):
mkdir(output_folder)
eval_result = inference(model, data_loader, dataset_name, device,
output_folder, **kwargs)
eval_results.append(eval_result)
print("\nTotal detection speed1: %.1f FPS" % (4952 / timer.toc()))
return eval_results
def run_demo(cfg, ckpt, score_threshold, images_dir, output_dir, onnx_dir, dataset_type):
if dataset_type == "voc":
class_names = VOCDataset.class_names
elif dataset_type == 'coco':
class_names = COCODataset.class_names
else:
raise NotImplementedError('Not implemented now.')
device = torch.device(cfg.MODEL.DEVICE)
device = "cpu" if not torch.cuda.is_available() else device
model = build_detection_model(cfg)
model = model.to(device)
checkpointer = CheckPointer(model, save_dir=cfg.OUTPUT_DIR)
checkpointer.load(ckpt, use_latest=ckpt is None)
weight_file = ckpt if ckpt else checkpointer.get_checkpoint_file()
print('Loaded weights from {}'.format(weight_file))
image_paths = glob.glob(os.path.join(images_dir, '*.jpg'))
mkdir(output_dir)
cpu_device = torch.device("cpu")
transforms = build_transforms(cfg, is_train=False)
model.eval()
# get model ready for onnx export
mkdir(onnx_dir)
model_onnx = build_detection_model(cfg)
model_onnx = model_onnx.to(device)
checkpointer_onnx = CheckPointer(model_onnx, save_dir=cfg.OUTPUT_DIR)
checkpointer_onnx.load(ckpt, use_latest=ckpt is None)
# replace the SSD box head postprocessor with the onnx version for exporting
model_onnx.box_head.post_processor = PostProcessorOnnx(cfg)
model_onnx.eval()
# export with ONNX
# onnx modle takes the name of the pth ckpt file
model_onnx_name = os.path.basename(ckpt).split('.')[0] + ".onnx"
model_onnx_path = os.path.join(onnx_dir, model_onnx_name)
if not os.path.exists(model_onnx_path):
print(f'Model exported as onnx to {model_onnx_path}')
dummy_input = torch.zeros(
[1, 3, cfg.INPUT.IMAGE_SIZE, cfg.INPUT.IMAGE_SIZE]).to(device)
torch.onnx.export(model_onnx,
dummy_input,
model_onnx_path,
export_params=True,
do_constant_folding=True,
opset_version=11,
input_names=['input'],
output_names=['boxes', 'scores', 'labels'],
dynamic_axes={
'input': {0: 'batch_size', 2: "height", 3: "width"}},
verbose=False)
# load exported onnx model for inference test
print(
f'Loading exported onnx model from {model_onnx_path} for inference comparison test')
onnx_runtime_sess = onnxruntime.InferenceSession(model_onnx_path)
for i, image_path in enumerate(image_paths):
start = time.time()
image_name = os.path.basename(image_path)
image = np.array(Image.open(image_path).convert("RGB"))
height, width = image.shape[:2]
images = transforms(image)[0].unsqueeze(0)
load_time = time.time() - start
start = time.time()
result = model(images.to(device))[0]
inference_time = time.time() - start
result = result.resize((width, height)).to(cpu_device).numpy()
boxes, labels, scores = result['boxes'], result['labels'], result['scores']
indices = scores > score_threshold
boxes, labels, scores = boxes[indices], labels[indices], scores[indices]
meters = ' | '.join(
[
'objects {:02d}'.format(len(boxes)),
'load {:03d}ms'.format(round(load_time * 1000)),
'inference {:03d}ms'.format(round(inference_time * 1000)),
'FPS {}'.format(round(1.0 / inference_time))
]
)
print('Pytorch: ({:04d}/{:04d}) {}: {}'.format(i + 1,
len(image_paths), image_name, meters))
drawn_image = draw_boxes(image, boxes, labels,
scores, class_names).astype(np.uint8)
Image.fromarray(drawn_image).save(
os.path.join(output_dir, "pytorch_" + image_name))
"""
Compute ONNX Runtime output prediction
"""
start = time.time()
ort_inputs = {onnx_runtime_sess.get_inputs()[0].name: np.array(images)}
boxes, scores, labels = onnx_runtime_sess.run(None, ort_inputs)
inference_time = time.time() - start
indices = scores > score_threshold
boxes, labels, scores = boxes[indices], labels[indices], scores[indices]
# resize bounding boxes to size of the original image
boxes[:, 0::2] *= (width)
boxes[:, 1::2] *= (height)
meters = ' | '.join(
[
'objects {:02d}'.format(len(boxes)),
'load {:03d}ms'.format(round(load_time * 1000)),
'inference {:03d}ms'.format(round(inference_time * 1000)),
'FPS {}'.format(round(1.0 / inference_time))
]
)
print('Onnx: ({:04d}/{:04d}) {}: {}'.format(i + 1,
len(image_paths),
image_name, meters))
drawn_image = draw_boxes(image, boxes, labels,
scores, class_names).astype(np.uint8)
Image.fromarray(drawn_image).save(
os.path.join(output_dir, "onnx_" + image_name))
def run_demo(cfg, ckpt, score_threshold, images_dir, output_dir):
class_names = VOCDataset.class_names
device = torch.device(cfg.MODEL.DEVICE)
model = build_detection_model(cfg)
model = model.to(device)
checkpointer = CheckPointer(model, save_dir=cfg.OUTPUT_DIR)
checkpointer.load(ckpt, use_latest=ckpt is None)
weight_file = ckpt if ckpt else checkpointer.get_checkpoint_file()
print('Loaded weights from {}'.format(weight_file))
image_paths = glob.glob(os.path.join(images_dir, '*.bmp'))
mkdir(output_dir)
cpu_device = torch.device("cpu")
transforms = build_transforms(cfg, is_train=False)
model.eval()
for i, image_path in enumerate(image_paths):
start = time.time()
image_name = os.path.basename(image_path)
image = np.array(Image.open(image_path).convert("RGB"))
height, width = image.shape[:2]
images = transforms(image)[0].unsqueeze(0)
load_time = time.time() - start
start = time.time()
result = model(images.to(device))[0]
inference_time = time.time() - start
result = result.resize((width, height)).to(cpu_device).numpy()
boxes, labels, scores = result['boxes'], result['labels'], result[
'scores']
indices = scores > score_threshold
boxes = boxes[indices]
labels = labels[indices]
meters = ' | '.join([
'objects {:02d}'.format(len(boxes)),
'load {:03d}ms'.format(round(load_time * 1000)),
'inference {:03d}ms'.format(round(inference_time * 1000)),
'FPS {}'.format(round(1.0 / inference_time))
])
print('({:04d}/{:04d}) {}: {}'.format(i + 1, len(image_paths),
image_name, meters))
text = ['__background__']
resDic = {}
for j in range(len(boxes)):
xmin = int(boxes[j, 0])
ymin = int(boxes[j, 1])
xmax = int(boxes[j, 2])
ymax = int(boxes[j, 3])
if labels[j] == 1:
xmin += 140
xmax -= 130
elif labels[j] == 2:
xmin += 130
elif labels[j] == 4:
xmin += 40
hight = ymax - ymin
width = xmax - xmin
cropImg = image[ymin:ymin + hight, xmin:xmin + width]
cropImg = local_threshold(cropImg)
boxes[j, 0] = xmin
boxes[j, 1] = ymin
boxes[j, 2] = xmax
boxes[j, 3] = ymax
text_tmp = crnnOcr(Image.fromarray(cropImg))
if labels[j] == 2:
text_tmp = re.sub('[^\x00-\xff]', '/', text_tmp)
text.append(text_tmp)
resDic[class_names[labels[j]]] = text_tmp
result = json.dumps(resDic, ensure_ascii=False)
print(result)
def main():
parser = argparse.ArgumentParser(
description='Single Shot MultiBox Detector Training With PyTorch')
parser.add_argument(
"--config-file",
default="",
metavar="FILE",
help="path to config file",
type=str,
)
parser.add_argument("--local_rank", type=int, default=0)
parser.add_argument('--log_step',
default=10,
type=int,
help='Print logs every log_step')
parser.add_argument('--save_step',
default=2500,
type=int,
help='Save checkpoint every save_step')
parser.add_argument(
'--eval_step',
default=2500,
type=int,
help='Evaluate dataset every eval_step, disabled when eval_step < 0')
parser.add_argument('--use_tensorboard', default=True, type=str2bool)
parser.add_argument(
"--skip-test",
dest="skip_test",
help="Do not test the final model",
action="store_true",
)
parser.add_argument(
"opts",
help="Modify config options using the command-line",
default=None,
nargs=argparse.REMAINDER,
)
args = parser.parse_args()
num_gpus = int(
os.environ["WORLD_SIZE"]) if "WORLD_SIZE" in os.environ else 1
args.distributed = num_gpus > 1
args.num_gpus = num_gpus
if torch.cuda.is_available():
# This flag allows you to enable the inbuilt cudnn auto-tuner to
# find the best algorithm to use for your hardware.
torch.backends.cudnn.benchmark = True
if args.distributed:
torch.cuda.set_device(args.local_rank)
torch.distributed.init_process_group(backend="nccl",
init_method="env://")
synchronize()
# Train distance regression network
train_distance_regr()
cfg.merge_from_file(args.config_file)
cfg.merge_from_list(args.opts)
cfg.freeze()
if cfg.OUTPUT_DIR:
mkdir(cfg.OUTPUT_DIR)
logger = setup_logger("SSD", dist_util.get_rank(), cfg.OUTPUT_DIR)
logger.info("Using {} GPUs".format(num_gpus))
logger.info(args)
logger.info("Loaded configuration file {}".format(args.config_file))
with open(args.config_file, "r") as cf:
config_str = "\n" + cf.read()
logger.info(config_str)
logger.info("Running with config:\n{}".format(cfg))
model = train(cfg, args)
if not args.skip_test:
logger.info('Start evaluating...')
torch.cuda.empty_cache() # speed up evaluating after training finished
do_evaluation(cfg, model, distributed=args.distributed)
def run_demo(cfg, ckpt, score_threshold, images_dir, output_dir, dataset_type):
if dataset_type == "voc":
class_names = VOCDataset.class_names
elif dataset_type == 'coco':
class_names = COCODataset.class_names
else:
raise NotImplementedError('Not implemented now.')
device = torch.device(cfg.MODEL.DEVICE)
smoke_name_dic = ('__background__', '一次性快餐盒', '书籍纸张', '充电宝', '剩饭剩菜', '包',
'垃圾桶', '塑料器皿', '塑料玩具', '塑料衣架', '大骨头', '干电池', '快递纸袋',
'插头电线', '旧衣服', '易拉罐', '枕头', '果皮果肉', '毛绒玩具', '污损塑料',
'污损用纸', '洗护用品', '烟蒂', '牙签', '玻璃器皿', '砧板', '筷子', '纸盒纸箱',
'花盆', '茶叶渣', '菜帮菜叶', '蛋壳', '调料瓶', '软膏', '过期药物', '酒瓶',
'金属厨具', '金属器皿', '金属食品罐', '锅', '陶瓷器皿', '鞋', '食用油桶', '饮料瓶',
'鱼骨')
model = build_detection_model(cfg)
cpu_device = torch.device("cpu")
model = model.to(device)
checkpointer = CheckPointer(model, save_dir=cfg.OUTPUT_DIR)
checkpointer.load(ckpt, use_latest=ckpt is None)
weight_file = ckpt if ckpt else checkpointer.get_checkpoint_file()
print('Loaded weights from {}'.format(weight_file))
image_paths = glob.glob(os.path.join(images_dir, '*.jpg'))
mkdir(output_dir)
transforms = build_transforms(cfg, is_train=False)
model.eval()
miss = 0
for i, image_path in enumerate(image_paths):
start = time.time()
image_name = os.path.basename(image_path)
cv_image = cv2.imread(image_path)
PIL_image = Image.open(image_path)
image = np.array(Image.open(image_path).convert("RGB"))
height, width = image.shape[:2]
images = transforms(image)[0].unsqueeze(0)
load_time = time.time() - start
start = time.time()
result = model(images.to(device))[0]
inference_time = time.time() - start
result = result.resize((width, height)).to(cpu_device).numpy()
boxes, labels, scores = result['boxes'], result['labels'], result[
'scores']
indices = scores > score_threshold
boxes = boxes[indices]
labels = labels[indices]
scores = scores[indices]
miss = miss + (1 - len(boxes))
meters = ' | '.join([
'objects {:02d}'.format(len(boxes)),
'load {:03d}ms'.format(round(load_time * 1000)),
'inference {:03d}ms'.format(round(inference_time * 1000)),
'FPS {}'.format(round(1.0 / inference_time))
])
print('({:04d}/{:04d}) {}: {}'.format(i + 1, len(image_paths),
image_name, meters))
draw_ = ImageDraw.Draw(PIL_image)
for c in range(len(scores)):
text = smoke_name_dic[labels[c]]
font = ImageFont.truetype(
'/usr/share/fonts/truetype/arphic/uming.ttc', 40)
draw_.text((int(boxes[c][0]) + 2, int(boxes[c][1]) - 2),
text, (255, 0, 0),
font=font)
cv_image = cv2.cvtColor(np.asarray(PIL_image), cv2.COLOR_RGB2BGR)
for c in range(len(scores)):
cv2.rectangle(cv_image, (int(boxes[c][0]), int(boxes[c][1])),
(int(boxes[c][2]), int(boxes[c][3])), (0, 0, 255), 4)
cv2.imwrite(os.path.join(output_dir, image_name), cv_image)
smoke_count = len(image_paths)
print("出现：%d 漏掉： %d 漏检率：%.2f" % (smoke_count, miss, miss / smoke_count))
def run_demo(cfg, model, score_threshold, images_dir, output_dir):
device = torch.device(cfg.MODEL.DEVICE)
class_names = VOCDataset.class_names
mkdir(output_dir)
cpu_device = torch.device("cpu")
transforms = build_transforms(cfg, is_train=False)
model.eval()
start = time.time()
image_name = os.path.basename(images_dir)
image = np.array(Image.open(images_dir).convert("RGB"))
height, width = image.shape[:2]
images = transforms(image)[0].unsqueeze(0)
load_time = time.time() - start
start = time.time()
result = model(images.to(device))[0]
inference_time = time.time() - start
result = result.resize((width, height)).to(cpu_device).numpy()
boxes, labels, scores = result['boxes'], result['labels'], result['scores']
indices = scores > score_threshold
boxes = boxes[indices]
labels = labels[indices]
meters = ' | '.join([
'objects {:02d}'.format(len(boxes)),
'load {:03d}ms'.format(round(load_time * 1000)),
'inference {:03d}ms'.format(round(inference_time * 1000)),
'FPS {}'.format(round(1.0 / inference_time))
])
print('({:04d}) {}: {}'.format(len(images_dir), image_name, meters))
text = ['__background__']
resDic = {}
for j in range(len(boxes)):
xmin = int(boxes[j, 0])
ymin = int(boxes[j, 1])
xmax = int(boxes[j, 2])
ymax = int(boxes[j, 3])
if labels[j] == 1:
xmin += 140
xmax -= 130
elif labels[j] == 2:
xmin += 130
elif labels[j] == 4:
xmin += 40
hight = ymax - ymin
width = xmax - xmin
cropImg = image[ymin:ymin + hight, xmin:xmin + width]
cropImg = local_threshold(cropImg)
text_tmp = crnnOcr(Image.fromarray(cropImg))
if labels[j] == 2:
text_tmp = re.sub('[^\x00-\xff]', '/', text_tmp)
text.append(text_tmp)
resDic[class_names[labels[j]]] = text_tmp
return json.dumps(resDic, ensure_ascii=False).encode('utf-8')
def main():
# 解析命令行 读取配置文件
'''
规定了模型的基本参数，训练的类，一共是20类加上背景所以是21
模型的输入大小，为了不对原图造成影响，一般是填充为300*300的图像
训练的文件夹路径2007和2012，测试的文件夹路径2007
最大迭代次数为120000.学习率还有gamma的值，总之就是一系列的超参数
输出的文件目录
MODEL:
NUM_CLASSES: 21
INPUT:
IMAGE_SIZE: 300
DATASETS:
TRAIN: ("voc_2007_trainval", "voc_2012_trainval")
TEST: ("voc_2007_test", )
SOLVER:
MAX_ITER: 120000
LR_STEPS: [80000, 100000]
GAMMA: 0.1
BATCH_SIZE: 32
LR: 1e-3
OUTPUT_DIR: 'outputs/vgg_ssd300_voc0712'
Returns:
'''
parser = argparse.ArgumentParser(description='Single Shot MultiBox Detector Training With PyTorch')
parser.add_argument(
"--config-file",
default="configs/vgg_ssd300_voc0712.yaml",
# default="configs/vgg_ssd300_visdrone0413.yaml",
metavar="FILE",
help="path to config file",
type=str,
)
# 每2500步保存一次文件，并且验证一次文件，记录是每10次记录一次，然后如果不想看tensor的记录的话，可以关闭，使用的是tensorboardX
parser.add_argument("--local_rank", type=int, default=0)
parser.add_argument('--log_step', default=10, type=int, help='Print logs every log_step')
parser.add_argument('--save_step', default=2500, type=int, help='Save checkpoint every save_step')
parser.add_argument('--eval_step', default=2500, type=int, help='Evaluate dataset every eval_step, disabled when eval_step < 0')
parser.add_argument('--use_tensorboard', default=True, type=str2bool)
parser.add_argument(
"--skip-test",
dest="skip_test",
help="Do not test the final model",
action="store_true",
)
parser.add_argument(
"opts",
help="Modify config options using the command-line",
default=None,
nargs=argparse.REMAINDER,
)
# 参数解析，可以使用多GPU进行训练
args = parser.parse_args()
num_gpus = int(os.environ["WORLD_SIZE"]) if "WORLD_SIZE" in os.environ else 1
args.distributed = num_gpus > 1
args.num_gpus = num_gpus
# 做一些启动前必要的检查
if torch.cuda.is_available():
# This flag allows you to enable the inbuilt cudnn auto-tuner to
# find the best algorithm to use for your hardware.
torch.backends.cudnn.benchmark = True
if args.distributed:
torch.cuda.set_device(args.local_rank)
torch.distributed.init_process_group(backend="nccl", init_method="env://")
synchronize()
cfg.merge_from_file(args.config_file)
cfg.merge_from_list(args.opts)
cfg.freeze()
# 创建模型输出文件夹
if cfg.OUTPUT_DIR:
mkdir(cfg.OUTPUT_DIR)
# 使用logger来进行记录
logger = setup_logger("SSD", dist_util.get_rank(), cfg.OUTPUT_DIR)
logger.info("Using {} GPUs".format(num_gpus))
logger.info(args)
# 加载配置文件
logger.info("Loaded configuration file {}".format(args.config_file))
with open(args.config_file, "r") as cf:
config_str = "\n" + cf.read()
logger.info(config_str)
logger.info("Running with config:\n{}".format(cfg))
# 模型训练
# model = train(cfg, args)
model = train(cfg, args)
# 开始进行验证
if not args.skip_test:
logger.info('Start evaluating...')
torch.cuda.empty_cache() # speed up evaluating after training finished
do_evaluation(cfg, model, distributed=args.distributed)
def run_demo(cfg, ckpt, score_threshold, images_dir, dataset_type):
if dataset_type == "voc":
class_names = VOCDataset.class_names
elif dataset_type == "pick":
class_names = PICKDataset.class_names
elif dataset_type == "cotb":
class_names = COTBDataset.class_names
elif dataset_type == 'coco':
class_names = COCODataset.class_names
else:
raise NotImplementedError('Not implemented now.')
device = torch.device(cfg.MODEL.DEVICE)
model = build_detection_model(cfg)
model = model.to(device)
checkpointer = CheckPointer(model, save_dir=cfg.OUTPUT_DIR)
checkpointer.load(ckpt, use_latest=ckpt is None)
weight_file = ckpt if ckpt else checkpointer.get_checkpoint_file()
print('Loaded weights from {}'.format(weight_file))
train_epoch = weight_file.split('/')[2]
train_epoch = train_epoch.split('.')[0].split('_')[1]
save_path = os.path.join('demo', dataset_type, cfg.MODEL.BACKBONE.NAME,
train_epoch)
image_paths = glob.glob(os.path.join(images_dir, '*.jpg')) + glob.glob(
os.path.join(images_dir, '*.jpeg'))
mkdir(save_path)
cpu_device = torch.device("cpu")
transforms = build_transforms(cfg, is_train=False)
model.eval()
for i, image_path in enumerate(image_paths):
start = time.time()
image_name = os.path.basename(image_path)
image = np.array(Image.open(image_path).convert("RGB"))
height, width = image.shape[:2]
images = transforms(image)[0].unsqueeze(0)
load_time = time.time() - start
start = time.time()
result = model(images.to(device))[0]
inference_time = time.time() - start
result = result.resize((width, height)).to(cpu_device).numpy()
boxes, labels, scores = result['boxes'], result['labels'], result[
'scores']
indices = scores > score_threshold
boxes = boxes[indices]
labels = labels[indices]
scores = scores[indices]
meters = ' | '.join([
'objects {:02d}'.format(len(boxes)),
'load {:03d}ms'.format(round(load_time * 1000)),
'inference {:03d}ms'.format(round(inference_time * 1000)),
'FPS {}'.format(round(1.0 / inference_time))
])
print('({:04d}/{:04d}) {}: {}'.format(i + 1, len(image_paths),
image_name, meters))
drawn_image = draw_boxes(image, boxes, labels, scores,
class_names).astype(np.uint8)
Image.fromarray(drawn_image).save(os.path.join(save_path, image_name))
def run_demo(cfg, ckpt, score_threshold, images_dir, output_dir, dataset_type, gen_heatmap):
if dataset_type == "voc":
class_names = VOCDataset.class_names
elif dataset_type == 'coco':
class_names = COCODataset.class_names
else:
raise NotImplementedError('Not implemented now.')
if torch.cuda.is_available():
device = torch.device(cfg.MODEL.DEVICE)
else:
device = torch.device("cpu")
model = build_detection_model(cfg)
model = model.to(device)
checkpointer = CheckPointer(model, save_dir=cfg.OUTPUT_DIR)
checkpointer.load(ckpt, use_latest=ckpt is None)
weight_file = ckpt if ckpt else checkpointer.get_checkpoint_file()
print('Loaded weights from {}'.format(weight_file))
mkdir(output_dir)
cpu_device = torch.device("cpu")
transforms = build_transforms(cfg, is_train=False)
model.eval()
dist_regr_model = DistanceRegrNet(2)
dist_regr_model = load_model_weight(dist_regr_model, device) # load weights
dist_regr_model.eval()
X_scaler = load_standardizer(Standardizer())
person_label_idx = class_names.index('person')
centroid_tracker = CentroidTracker()
capture = cv2.VideoCapture(0)
while capture.isOpened():
ret, frame = capture.read()
single_frame_render_time = 0
if ret:
image = frame
height, width = image.shape[:2]
start_time = time.time()
images = transforms(frame)[0].unsqueeze(0)
result = model(images.to(device))[0]
result = result.resize((width, height)).to(cpu_device).numpy()
single_frame_render_time += round((time.time() - start_time) * 1000, 3)
print(f"MobileNet SSD Inference time {round((time.time() - start_time) * 1000, 3)}ms")
boxes, labels, scores = result['boxes'], result['labels'], result['scores']
# remove all non person class detections
indices = np.logical_and(scores > score_threshold,
labels == person_label_idx)
boxes = boxes[indices]
labels = labels[indices]
scores = scores[indices]
distances = None
# create gaussian mixture models and kde plots only if centers detected
if len(boxes) != 0:
centers = np.apply_along_axis(get_mid_point, 1, boxes)
image = draw_points(image, centers) # draw center points on image
# Distance Regression
start_time = time.time()
# As boxes is in (xmin, ymin, xmax, ymax) format
# X should always have width, height format
width = boxes[:, 2] - boxes[:, 0]
height = boxes[:, 3] - boxes[:, 1]
X = np.column_stack((width, height))
X_scaled = X_scaler.transform(X)
distances = dist_regr_model(torch.Tensor(X_scaled).to(device)).to(cpu_device).numpy()
single_frame_render_time += round((time.time() - start_time) * 1000, 3)
print(f"Distance Regression Inference time {round(time.time() - start_time, 4) * 1000}ms")
# object tracking with centroids
start_time = time.time()
objects = centroid_tracker.update(centers, distances)
# loop over the tracked objects
# for (objectID, centroid) in objects.items():
# print("Center Distances tracked overtime")
# print(centroid_tracker.obj_distance_counts[objectID])
single_frame_render_time += round((time.time() - start_time) * 1000, 3)
print(f"Centroid Tracking Update time {round(time.time() - start_time, 4) * 1000}ms")
if len(centers) > 1:
# reset center point ranges to a min of 0 and max of 100
_x = centers[:, 0]
_y = centers[:, 1]
centers[:, 0] = reset_range(max(_x), min(_x), 100, 0, _x)
centers[:, 1] = reset_range(max(_y), min(_y), 100, 0, _y)
# DBSCAN Clustering
start_time = time.time()
dbscan_center = DBSCAN(eps=18)
dbscan_center.fit(centers)
# print("DBSCAN Clusters", dbscan_center._labels)
# print("Unique number of clusters", len(set(dbscan_center._labels)))
single_frame_render_time += round((time.time() - start_time) * 1000, 3)
print(f"DBSCAN Clustering time {round((time.time() - start_time) * 1000, 3)}ms")
if gen_heatmap:
image = generate_cv2_heatmap(centers, dbscan_center._labels, None, None,
len(set(dbscan_center._labels)),
covariance_type='diag')
cv2.imshow("frame", image)
if not gen_heatmap:
drawn_image = draw_boxes(image, boxes, labels, scores, distances, class_names).astype(np.uint8)
cv2.imshow("frame", drawn_image)
print(f"Total time to render one frame {single_frame_render_time}." +
f"FPS {round(1 / (single_frame_render_time / 1000))}")
key = cv2.waitKey(1)
if key & 0xFF == ord('x'):
break
else:
break
print("Distance counts for tracked objects")
print(centroid_tracker.obj_distance_counts)
write_file = f'{output_dir}/dist_regr_results/{round(time.time())}.txt'
print(f"Writing the distance values to file {write_file}")
os.makedirs(f'{output_dir}/dist_regr_results', exist_ok=True)
with open(write_file, 'w') as fw:
for key, arr in centroid_tracker.obj_distance_counts.items():
arr = [str(v) for v in arr]
fw.write(str(key) + ',' + ','.join(arr))
fw.write('\n')
capture.release()
cv2.destroyAllWindows()
def run_demo(cfg,
ckpt,
score_threshold,
images_dir,
output_dir,
dataset_type,
model_path=None):
if dataset_type == "voc":
class_names = VOCDataset.class_names
elif dataset_type == 'coco':
class_names = COCODataset.class_names
else:
class_names = TxtDataset(dataset_name=dataset_type).class_names
# else:
# raise NotImplementedError('Not implemented now.')
device = torch.device(cfg.MODEL.DEVICE)
model = build_detection_model(cfg)
model = model.to(device)
checkpointer = CheckPointer(model, save_dir=cfg.OUTPUT_DIR)
checkpointer.load(ckpt, use_latest=ckpt is None)
##
# model.backbone.bn_fuse()#需要修改demo.py 要bn_fuse 因为fpga端没有bn
# model.to(device)
# ##
if model_path is None:
checkpointer.load(ckpt, use_latest=ckpt is None)
weight_file = ckpt if ckpt else checkpointer.get_checkpoint_file()
print('Loaded weights from {}'.format(weight_file))
else:
model.load_state_dict(torch.load(model_path))
if cfg.TEST.BN_FUSE is True:
print('BN_FUSE.')
model.backbone.bn_fuse()
model.to(device)
image_paths = glob.glob(os.path.join(images_dir, '*.jpg')) #.png
mkdir(output_dir)
cpu_device = torch.device("cpu")
transforms = build_transforms(cfg, is_train=False)
model.eval()
for i, image_path in enumerate(image_paths):
start = time.time()
image_name = os.path.basename(image_path)
image = np.array(Image.open(image_path).convert("RGB"))
height, width = image.shape[:2]
images = transforms(image)[0].unsqueeze(0)
load_time = time.time() - start
start = time.time()
result = model(images.to(device))[0]
inference_time = time.time() - start
result = result.resize((width, height)).to(cpu_device).numpy()
boxes, labels, scores = result['boxes'], result['labels'], result[
'scores']
indices = scores > score_threshold
boxes = boxes[indices]
labels = labels[indices]
scores = scores[indices]
meters = ' | '.join([
'objects {:02d}'.format(len(boxes)),
'load {:03d}ms'.format(round(load_time * 1000)),
'inference {:03d}ms'.format(round(inference_time * 1000)),
'FPS {}'.format(round(1.0 / inference_time))
])
print('({:04d}/{:04d}) {}: {}'.format(i + 1, len(image_paths),
image_name, meters))
drawn_image = draw_boxes(image, boxes, labels, scores,
class_names).astype(np.uint8)
Image.fromarray(drawn_image).save(os.path.join(output_dir, image_name))
def run_demo(cfg, ckpt, score_threshold, images_dir, output_dir, dataset_type):
if dataset_type == "voc":
class_names = VOCDataset.class_names
elif dataset_type == 'coco':
class_names = COCODataset.class_names
else:
raise NotImplementedError('Not implemented now.')
if torch.cuda.is_available():
device = torch.device(cfg.MODEL.DEVICE)
else:
device = torch.device("cpu")
model = build_detection_model(cfg)
model = model.to(device)
checkpointer = CheckPointer(model, save_dir=cfg.OUTPUT_DIR)
checkpointer.load(ckpt, use_latest=ckpt is None)
weight_file = ckpt if ckpt else checkpointer.get_checkpoint_file()
print('Loaded weights from {}'.format(weight_file))
images_dir = 'datasets/MOT16/train/MOT16-02/img1'
image_paths = sorted(glob.glob(os.path.join(images_dir, '*.jpg')))
mkdir(output_dir)
cpu_device = torch.device("cpu")
transforms = build_transforms(cfg, is_train=False)
model.eval()
person_label_idx = class_names.index('person')
centroid_tracker = CentroidTracker()
wfile = open('py-motmetrics/motmetrics/data/MOT16/predicted/MOT16-02.txt',
'w')
inference_times = []
for i, image_path in enumerate(image_paths):
image_name = os.path.basename(image_path)
start_time = time.time()
image = np.array(Image.open(image_path).convert("RGB"))
height, width = image.shape[:2]
images = transforms(image)[0].unsqueeze(0)
result = model(images.to(device))[0]
result = result.resize((width, height)).to(cpu_device).numpy()
boxes, labels, scores = result['boxes'], result['labels'], result[
'scores']
# remove all non person class detections
indices = np.logical_and(scores > score_threshold,
labels == person_label_idx)
boxes = boxes[indices]
distances = None
inference_times.append(time.time() - start_time)
print(time.time() - start_time)
if len(boxes) != 0:
centers = np.apply_along_axis(get_mid_point, 1, boxes)
# object tracking with centroids
centroid_tracker.update(centers, distances, boxes)
fnum = int(image_name.split('.')[0])
# loop over the tracked objects
for (objID, bbox_) in centroid_tracker.obj_bbox.items():
xm, ym = bbox_[0], bbox_[1]
w, h = bbox_[2] - bbox_[0], bbox_[3] - bbox_[1]
output = f"{fnum},{objID},{xm},{ym},{w},{h},-1,-1,-1\n"
wfile.write(output)
# drawn_image = draw_boxes(image, boxes, labels, scores, distances, class_names).astype(np.uint8)
# Image.fromarray(drawn_image).save(os.path.join(output_dir, image_name))
framerates = [1 / tm for tm in inference_times]
print(
f"Avg frame rate is {sum(framerates) / len(framerates)} for {len(framerates)} frames"
)
wfile.close()
def run_demo(cfg, ckpt, score_threshold, images_dir, output_dir, dataset_type):
if dataset_type == "voc":
class_names = VOCDataset.class_names
elif dataset_type == 'coco':
class_names = COCODataset.class_names
else:
raise NotImplementedError('Not implemented now.')
device = torch.device(cfg.MODEL.DEVICE)
model = build_detection_model(cfg)
model = model.to(device)
checkpointer = CheckPointer(model, save_dir=cfg.OUTPUT_DIR)
checkpointer.load(ckpt, use_latest=ckpt is None)
weight_file = ckpt if ckpt else checkpointer.get_checkpoint_file()
print('Loaded weights from {}'.format(weight_file))
image_paths = glob.glob(os.path.join(images_dir, '*.jpg'))
mkdir(output_dir)
cpu_device = torch.device("cpu")
transforms = build_transforms(cfg, is_train=False)
model.eval()
_t = {'im_detect': Timer()}
timer = Timer()
timer.tic()
inference_time_list=[]
load_time_list = []
for image_path in image_paths:
start = time.time()
image_name = os.path.basename(image_path)
image = np.array(Image.open(image_path).convert("RGB"))
height, width = image.shape[:2]
images = transforms(image)[0].unsqueeze(0)
load_time = time.time() - start
load_time_list.append(1000*load_time)
_t['im_detect'].tic()
#start = time.time()
#print('1')
result = model(images.to(device))[0]
#print('2')
result = result.resize((width, height)).to(cpu_device).numpy()
boxes, labels, scores = result['boxes'], result['labels'], result['scores']
indices = scores > score_threshold
boxes = boxes[indices]
labels = labels[indices]
scores = scores[indices]
#inference_time = time.time() - start
inference_time = _t['im_detect'].toc()
#print(1000*(inference_time))
inference_time_list.append(1000*inference_time)
meters = ' | '.join(
[
'objects {:02d}'.format(len(boxes)),
'load {:03d}ms'.format(round(load_time * 1000)),
'inference {:03d}ms'.format(round(inference_time * 1000)),
'FPS {}'.format(round(1.0 / inference_time))
]
)
# print('({:04d}/{:04d}) {}: {}'.format(i + 1, len(image_paths), image_name, meters))
#drawn_image = draw_boxes(image, boxes, labels, scores, class_names).astype(np.uint8)
#Image.fromarray(drawn_image).save(os.path.join(output_dir, image_name))
_t['im_detect'].clear()
N = len(inference_time_list)//2
total_time_list = np.array(inference_time_list) + np.array(load_time_list)
total_time_list.sort()
inference_time_list.sort()
det_time = np.mean(total_time_list[:N])#/BATCH_SIZE
best_det_time = np.min(total_time_list)#/BATCH_SIZE
print("Total test time: %.2f s" % (timer.toc()))
print("\nTotal detection speed: %.1f FPS" % (len(inference_time_list)/timer.toc()))
print("\nAvg detection speed: %.1f FPS" % (1000./det_time))
print("Best detection speed: %.1f FPS" % (1000./best_det_time))
def run_demo(cfg, ckpt, score_threshold, images_dir, output_dir, dataset_type):
if dataset_type == "voc":
class_names = VOCDataset.class_names
elif dataset_type == 'coco':
class_names = COCODataset.class_names
else:
raise NotImplementedError('Not implemented now.')
device = torch.device(cfg.MODEL.DEVICE)
cpu_device = torch.device("cpu")
model = build_detection_model(cfg)
model = model.to(cpu_device)
checkpointer = CheckPointer(model, save_dir=cfg.OUTPUT_DIR)
checkpointer.load(ckpt, use_latest=ckpt is None)
weight_file = ckpt if ckpt else checkpointer.get_checkpoint_file()
print('Loaded weights from {}'.format(weight_file))
image_paths = glob.glob(os.path.join(images_dir, '*.jpg'))
mkdir(output_dir)
transforms = build_transforms(cfg, is_train=False)
model.eval()
for i, image_path in enumerate(image_paths):
start = time.time()
image_name = os.path.basename(image_path)
image = np.array(Image.open(image_path).convert("RGB"))
height, width = image.shape[:2]
images = transforms(image)[0].unsqueeze(0)
load_time = time.time() - start
start = time.time()
result = model(images.to(cpu_device))[0]
inference_time = time.time() - start
result = result.resize((width, height)).numpy()
boxes, labels, scores = result['boxes'], result['labels'], result[
'scores']
indices = scores > score_threshold
boxes = boxes[indices]
labels = labels[indices]
scores = scores[indices]
meters = ' | '.join([
'objects {:02d}'.format(len(boxes)),
'load {:03d}ms'.format(round(load_time * 1000)),
'inference {:03d}ms'.format(round(inference_time * 1000)),
'FPS {}'.format(round(1.0 / inference_time))
])
print('({:04d}/{:04d}) {}: {}'.format(i + 1, len(image_paths),
image_name, meters))
for i in range(len(labels)):
text = str(label_name[labels[i]]) + str(round(scores[i], 2))
cv2.rectangle(image, tuple(boxes[i][:2]), tuple(boxes[i][2:]),
color, 3)
image = Image.fromarray(image)
draw = ImageDraw.Draw(image)
draw.text(tuple([boxes[i][0], boxes[i][1] - 40]),
text,
color,
font=fontStyle)
image = np.asarray(image)
cv2.imshow('drawn_image', image)
# drawn_image = draw_boxes(image, boxes, labels, scores, class_names).astype(np.uint8)
Image.fromarray(image).save(os.path.join(output_dir, image_name))
def run_demo(cfg, ckpt, score_threshold, images_dir, output_dir, dataset_type,
gen_heatmap):
if dataset_type == "voc":
class_names = VOCDataset.class_names
elif dataset_type == 'coco':
class_names = COCODataset.class_names
else:
raise NotImplementedError('Not implemented now.')
if torch.cuda.is_available():
device = torch.device(cfg.MODEL.DEVICE)
else:
device = torch.device("cpu")
model = build_detection_model(cfg)
model = model.to(device)
checkpointer = CheckPointer(model, save_dir=cfg.OUTPUT_DIR)
checkpointer.load(ckpt, use_latest=ckpt is None)
weight_file = ckpt if ckpt else checkpointer.get_checkpoint_file()
print('Loaded weights from {}'.format(weight_file))
image_paths = glob.glob(os.path.join(images_dir, '*.jpg'))
mkdir(output_dir)
cpu_device = torch.device("cpu")
transforms = build_transforms(cfg, is_train=False)
model.eval()
dist_regr_model = DistanceRegrNet(2)
dist_regr_model = load_model_weight(dist_regr_model,
device) # load weights
dist_regr_model.eval()
X_scaler = load_standardizer(Standardizer())
person_label_idx = class_names.index('person')
for i, image_path in enumerate(image_paths):
start = time.time()
image_name = os.path.basename(image_path)
image = np.array(Image.open(image_path).convert("RGB"))
height, width = image.shape[:2]
images = transforms(image)[0].unsqueeze(0)
load_time = time.time() - start
start = time.time()
result = model(images.to(device))[0]
inference_time = time.time() - start
result = result.resize((width, height)).to(cpu_device).numpy()
boxes, labels, scores = result['boxes'], result['labels'], result[
'scores']
# remove all non person class detections
indices = np.logical_and(scores > score_threshold,
labels == person_label_idx)
boxes = boxes[indices]
labels = labels[indices]
scores = scores[indices]
distances = None
# create gaussian mixture models and kde plots only if centers detected
if len(boxes) != 0:
centers = np.apply_along_axis(get_mid_point, 1, boxes)
image = draw_points(image, centers) # draw center points on image
# reset center point ranges to a min of 0 and max of 100
_x = centers[:, 0]
_y = centers[:, 1]
centers[:, 0] = reset_range(max(_x), min(_x), 100, 0, _x)
centers[:, 1] = reset_range(max(_y), min(_y), 100, 0, _y)
# DBSCAN Clustering
start = time.time()
dbscan_center = DBSCAN(eps=18)
dbscan_center.fit(centers)
# print("dbscan clusters", dbscan_center._labels)
# print("Unique number of clusters", len(set(dbscan_center._labels)))
print(
f"DBSCAN clustering time {round((time.time() - start) * 1000, 3)}ms"
)
# Distance Regression
start_time = time.time()
# As boxes is in (xmin, ymin, xmax, ymax) format
# X should always have width, height format
width = boxes[:, 2] - boxes[:, 0]
height = boxes[:, 3] - boxes[:, 1]
X = np.column_stack((width, height))
X_scaled = X_scaler.transform(X)
distances = dist_regr_model(torch.Tensor(X_scaled).to(device))
print(
f"Distance Regr Inference time {round(time.time() - start_time, 4) * 1000}ms"
)
if gen_heatmap:
generate_sns_kde_heatmap(centers[:, 0], centers[:, 1], i,
image_name)
generate_sk_gaussian_mixture(centers,
dbscan_center._labels,
i,
image_name,
len(set(dbscan_center._labels)),
covariance_type='diag')
generate_cv2_heatmap(centers,
dbscan_center._labels,
i,
image_name,
len(set(dbscan_center._labels)),
covariance_type='diag')
meters = ' | '.join([
'objects {:02d}'.format(len(boxes)),
'load {:03d}ms'.format(round(load_time * 1000)),
'inference {:03d}ms'.format(round(inference_time * 1000)),
'FPS {}'.format(round(1.0 / inference_time))
])
print('({:04d}/{:04d}) {}: {}'.format(i + 1, len(image_paths),
image_name, meters))
drawn_image = draw_boxes(image, boxes, labels, scores, distances,
class_names).astype(np.uint8)
Image.fromarray(drawn_image).save(os.path.join(output_dir, image_name))
