Modules and utils for YOLOv5

class yolort.v5.AutoShape(model)

YOLOv5 input-robust model wrapper for passing cv2/np/PIL/torch inputs. Includes preprocessing, inference and NMS

classes = None

conf = 0.25

forward(imgs, size=640, augment=False, profile=False)

Inference from various sources. For height=640, width=1280, RGB images example inputs are:

• file: imgs = ‘data/images/zidane.jpg’ # str or PosixPath

• URI: = ‘https://ultralytics.com/images/zidane.jpg’

• OpenCV: = cv2.imread(‘image.jpg’)[:,:,::-1] # HWC BGR to RGB x(640,1280,3)

• PIL: = Image.open(‘image.jpg’) or ImageGrab.grab() # HWC x(640,1280,3)

• numpy: = np.zeros((640,1280,3)) # HWC

• torch: = torch.zeros(16,3,320,640) # BCHW (scaled to size=640, 0-1 values)

• multiple: = [Image.open(‘image1.jpg’), Image.open(‘image2.jpg’), …] # list of images

iou = 0.45

max_det = 1000

multi_label = False

class yolort.v5.Bottleneck(c1, c2, shortcut=True, g=1, e=0.5, version='r4.0')

Standard bottleneck

Parameters:

• c1 (int) – ch_in

• c2 (int) – ch_out

• shortcut (bool) – shortcut

• g (int) – groups

• e (float) – expansion

• version (str) – Module version released by ultralytics. Possible values are [“r3.1”, “r4.0”]. Default: “r4.0”.

class yolort.v5.BottleneckCSP(c1, c2, n=1, shortcut=True, g=1, e=0.5)

CSP Bottleneck https://github.com/WongKinYiu/CrossStagePartialNetworks

Parameters:

• c1 (int) – ch_in

• c2 (int) – ch_out

• n (int) – number

• shortcut (bool) – shortcut

• g (int) – groups

• e (float) – expansion

class yolort.v5.C3(c1, c2, n=1, shortcut=True, g=1, e=0.5, version='r4.0')

CSP Bottleneck with 3 convolutions

Parameters:

• c1 (int) – ch_in

• c2 (int) – ch_out

• n (int) – number

• shortcut (bool) – shortcut

• g (int) – groups

• e (float) – expansion

• version (str) – Module version released by ultralytics. Possible values are [“r4.0”]. Default: “r4.0”.

class yolort.v5.C3TR(c1, c2, n=1, shortcut=True, g=1, e=0.5)

class yolort.v5.Concat(dimension: int = 1)

class yolort.v5.Contract(gain=2)

class yolort.v5.Conv(c1, c2, k=1, s=1, p=None, g=1, act=True, version='r4.0')

Standard convolution

Parameters:

• c1 (int) – ch_in

• c2 (int) – ch_out

• k (int) – kernel

• s (int) – stride

• p (Optional[int]) – padding

• g (int) – groups

• act (bool or nn.Module) – determine the activation function

• version (str) – Module version released by ultralytics. Possible values are [“r3.1”, “r4.0”]. Default: “r4.0”.

forward_fuse(x: Tensor) → Tensor

class yolort.v5.DWConv(c1, c2, k=1, s=1, act=True, version='r4.0')

Depth-wise convolution class.

Parameters:

• c1 (int) – ch_in

• c2 (int) – ch_out

• k (int) – kernel

• s (int) – stride

• act (bool or nn.Module) – determine the activation function

• version (str) – Module version released by ultralytics. Possible values are [“r3.1”, “r4.0”]. Default: “r4.0”.

class yolort.v5.Detect(nc=80, anchors=(), ch=(), inplace=True)

onnx_dynamic = False

stride = None

class yolort.v5.Ensemble

class yolort.v5.Expand(gain=2)

class yolort.v5.Focus(c1, c2, k=1, s=1, p=None, g=1, act=True, version='r4.0')

Focus wh information into c-space

Parameters:

• c1 (int) – ch_in

• c2 (int) – ch_out

• k (int) – kernel

• s (int) – stride

• p (Optional[int]) – padding

• g (int) – groups

• act (bool or nn.Module) – determine the activation function

• version (str) – Module version released by ultralytics. Possible values are [“r3.1”, “r4.0”]. Default: “r4.0”.

class yolort.v5.GhostBottleneck(c1, c2, k=3, s=1)

class yolort.v5.GhostConv(c1, c2, k=1, s=1, g=1, act=True)

class yolort.v5.Model(cfg='yolov5s.yaml', ch=3, nc=None, anchors=None)

fuse()

info(verbose=False, img_size=640)

class yolort.v5.SPP(c1, c2, k=(5, 9, 13), version='r4.0')

class yolort.v5.SPPF(c1, c2, k=5, version='r4.0')

Spatial Pyramid Pooling - Fast (SPPF) layer for YOLOv5 by Glenn Jocher

yolort.v5.add_yolov5_context()

Temporarily add yolov5 folder to sys.path. Adapted from https://github.com/fcakyon/yolov5-pip/blob/0d03de6/yolov5/utils/general.py#L739-L754

torch.hub handles it in the same way: https://github.com/pytorch/pytorch/blob/d3e36fa/torch/hub.py#L387-L416

yolort.v5.attempt_download(file, repo='ultralytics/yolov5', hash_prefix=None)

yolort.v5.focus_transform(x: Tensor) → Tensor

x(b,c,w,h) -> y(b,4c,w/2,h/2)

yolort.v5.get_yolov5_size(depth_multiple, width_multiple)

yolort.v5.intersect_dicts(dict1, dict2, exclude=())

Dictionary intersection of matching keys and shapes, omitting ‘exclude’ keys, using dict1 values

yolort.v5.letterbox(im: ndarray, new_shape: Tuple[int, int] = (640, 640), color: Tuple[int, int, int] = (114, 114, 114), auto: bool = True, scale_fill: bool = False, scaleup: bool = True, stride: int = 32)

yolort.v5.load_yolov5_model(checkpoint_path: str, fuse: bool = False)

Creates a specified YOLOv5 model.

Note

Currently this tool is mainly used to load the checkpoints trained by yolov5 with support for versions v3.1, v4.0 (v5.0) and v6.0 (v6.1). In addition it is available for inference with AutoShape attached for versions v6.0 (v6.1).

Parameters:

• checkpoint_path (str) – path of the YOLOv5 model, i.e. ‘yolov5s.pt’

• fuse (bool) – fuse model Conv2d() + BatchNorm2d() layers. Default: False

Returns:

YOLOv5 pytorch model

yolort.v5.non_max_suppression(prediction, conf_thres=0.25, iou_thres=0.45, classes=None, agnostic=False, multi_label=False, labels=(), max_det=300)

Runs Non-Maximum Suppression (NMS) on inference results

Returns:

list of detections, on (n,6) tensor per image [xyxy, conf, cls]

yolort.v5.scale_coords(img1_shape, coords, img0_shape, ratio_pad=None)

yolort.v5.select_device(device='', batch_size=None, newline=True)

yolort.v5.set_logging(name=None, verbose=True)

yolort.v5.space_to_depth(x: Tensor) → Tensor

x(b,c,w,h) -> y(b,4c,w/2,h/2)