Horizontally split a single image with overlapping regions. Args: image: The image to split (H, W, C). split_width: Width of

Gather samples into batches for inference purposes Args: samples: list of samples of shape (C, H, W) Returns:

Encode a list of gts sequences into a np array and gives the corresponding* sequence lengths. Args: gts: list of ground-t

Build the target, and it's mask to be used from loss computation. Args: target: target coming from dataset output_sha

Build the target, and it's mask to be used from loss computation. Args: target: target coming from dataset output_sha

Compute the distance for each point of the map (xs, ys) to the (a, b) segment Args: xs : map of x coordinates (height, width)

Compute categorical cross-entropy loss for the model. Sequences are masked after the EOS character. Args: model_output: p

Compute CTC loss for the model. Args: model_output: predicted logits of the model target: list of target strings

Compute categorical cross-entropy loss for the model. Sequences are masked after the EOS character. Args: gt: the encoded

Compute CTC loss for the model. Args: model_output: predicted logits of the model gt: ground truth tensor

Compute categorical cross-entropy loss for the model. Sequences are masked after the EOS character. Args: model_output: p

Compute linknet loss, BCE with boosted box edges or focal loss. Focal loss implementation based on <https://github.com/tensorflow/addons/>`_.

Compute fast loss, 2 x Dice loss where the text kernel loss is scaled by 0.5. Args: out_map: output feature map of the model of s

Compute a batch of gts, masks, thresh_gts, thresh_masks from a list of boxes and a list of masks for each image. From there it computes the lo

Creates a random shadow mask Args: target_shape: the target shape (H, W) min_base_width: the relative minimum shadow base width

Crop localization boxes Args: boxes: ndarray of shape (N, 4) in relative or abs coordinates crop_box: box (xmin, ymin, xmax, ymax

Implements best path decoding as shown by Graves (Dissertation, p63), highly inspired from <https://github.com/githubharald/CTCDecoder>`_.

Implements best path decoding as shown by Graves (Dissertation, p63), highly inspired from <https://github.com/githubharald/CTCDecoder>`_.

Decode function for prediction Args: encoded: input tensor Returns: A tuple of torch.Tensor: predictions, lo

Creates a LinkNet decoder block

Performs dilation on a given tensor Args: x: boolean tensor of shape (N, C, H, W) kernel_size: the size of the kernel to use for

Draw a polygon threshold map on a canvas, as described in the DB paper Args: polygon : array of coord., to draw the boundary of t

Performs erosion on a given tensor Args: x: boolean tensor of shape (N, C, H, W) kernel_size: the size of the kernel to use for e

Takes a batch of rotated previously ORIENTED polys (N, 4, 2) (rectified by the classifier) and return the estimated angle ccw in degrees

Export the document as XML (hOCR-format) Args: **kwargs: additional keyword arguments passed to the Page.export_as_xml method

Format polygons into an array Args: polygons: the bounding boxes use_polygons: whether polygons should be considered

Forward an image through the predictor Args: predictor: instance of OCRPredictor image: image to process device: torch.de

Instantiate & load a pretrained model from HF hub. >>> from doctr.models import from_hub >>> model = from_hub("mindee/fasterrcnn_mobilenet_v3

Load pretrained parameters onto the model Args: path_or_url: the path or URL to the model parameters (checkpoint) **k

Load pretrained parameters onto the model Args: path_or_url: the path or URL to the model parameters (checkpoint) **k

Load pretrained parameters onto the model Args: path_or_url: the path or URL to the model parameters (checkpoint) **k

Load pretrained parameters onto the model Args: path_or_url: the path or URL to the model parameters (checkpoint) **k

Load pretrained parameters onto the model Args: path_or_url: the path or URL to the model parameters (checkpoint) **k

Load pretrained parameters onto the model Args: path_or_url: the path or URL to the model parameters (checkpoint) **k

Load pretrained parameters onto the model Args: path_or_url: the path or URL to the model parameters (checkpoint) **k

Load pretrained parameters onto the model Args: path_or_url: the path or URL to the model parameters (checkpoint) **k

Load pretrained parameters onto the model Args: path_or_url: the path or URL to the model parameters (checkpoint) **k

Load pretrained parameters onto the model Args: path_or_url: the path or URL to the model parameters (checkpoint) **k

Interpret a web page as a PDF document >>> from doctr.io import DocumentFile >>> doc = DocumentFile.from_url("https://www.yoursite.co

Generate num_colors color for matplotlib Args: num_colors: number of colors to generate Returns: colors: list of generated c

This will return a list of libcudnn.so; it's hard to tell which one is being used

Compute the learnable position encoding via depthwise convolution. Args: x: A float tensor of shape (b, n, c).

Collects environment information for debugging purposes Returns: str: environment information

Slice an image into windows of shape (h_sp, w_sp). Args: img: A float tensor of shape (b, c, h, w). h_sp: Th

100 % borrowed from: https://github.com/huggingface/transformers/blob/main/src/transformers/models/vit/modeling_vit.py This method al

Load a predictor from doctr.models Args: det_arch: detection architecture reco_arch: recognition architecture assume_stra

Build a streamlit layout

* modified to run programmatically