Positionally encode points that are normalized to [0,1].

There are a lot of things that can get fucked up, so the simplest way to deal with potential problems is to collect the cv results into a sep

Expects a numpy array of length 2 in the final dimension. Requires the original image size in (H, W) format.

If plans_file_or_dict or dataset_json_file_or_dict are None, we will look for them in input_folder

Computes the stability score for a batch of masks. The stability score is the IoU between the binary masks obtained by thresholding th

does not rely on dataset.json

Determines nnUNet postprocessing. Its output is a postprocessing.pkl file in folder_predictions which can be used with apply_postprocessing_t

Converts all `BatchNorm2d` and `SyncBatchNorm` layers of provided module into `FrozenBatchNorm2d`. If `module` is itself an instance of eithe

configurations are unique within each plans file but differnet plans file can have configurations with the same name. In order to dis

Returns the positional encoding used to encode point prompts, applied to a dense set of points the shape of the image encoding.

Fetch model config from builtin (local library) configs.

this is the same as get_pool_and_conv_props_v2 from old nnunet :param spacing: :param patch_size: :param min_feature_map_size: min e

net_output must be (b, c, x, y(, z))) gt must be a label map (shape (b, 1, x, y(, z)) OR shape (b, x, y(, z))) or one hot encoding (b, c, x,

lock modules Args: unlocked_groups (int): leave last n layer groups unlocked (default: 0)

Resets the currently set image.

This function is specific for the default architecture in nnU-Net. If you change the architecture, there are chances you need to chan

Weights for CE and Dice do not need to sum to one. You can set whatever you want. :param soft_dice_kwargs: :param ce_kwargs:

Fetch model config from HuggingFace Hub.

Expects a torch tensor with shape Bx4. Requires the original image size in (H, W) format.

Expects a torch tensor with length 2 in the last dimension. Requires the original image size in (H, W) format.

logits has to have shape (c, x, y(, z)) where c is the number of classes/regions

Calculates boxes in XYXY format around masks. Return [0,0,0,0] for an empty mask. For input shape C1xC2x...xHxW, the output shape is C1xC2x

his is where you build the architecture according to the plans. There is no obligation to use get_network_from_plans, this is just a

Returns list of utf-8 byte and a corresponding list of unicode strings. The reversible bpe codes work on unicode strings. This means you

Calculate the combined loss expression for optimization of lambda

Returns canonicalized `text` (lowercase and punctuation removed). From: https://github.com/google-research/big_vision/blob/53f18caf27a9419231bbf0

used to collate default train_step and validation_step outputs. If you want something different then you gotta extend this we expect out

assumes that inference_nonlinearity was already applied! predicted_probabilities has to have shape (c, x, y(, z)) where c is the num

Convert applicable model parameters to low-precision (bf16 or fp16)