refer to: python PIL draw multiline text on image: https://stackoverflow.com/a/7698300/395857

TODO: add document here.

Initialize the weights in backbone. Args: pretrained (str, optional): Path to pre-trained weights. Defaults to No

Return the median value over the data in given window_size.

implemented ["wti", "ti"] interactions Args: sequence_output: repeat by batch or row, only sequence_output[0]

Returns the names of available CLIP models

Chooses static padding if you have specified an image size, and dynamic padding otherwise. Static padding is necessary for ONNX exporting of mo

:param encoder_output: bsz, seq_length, hidden :param decoder_output: :return:

Args: package_name(str): The name of package. Returns: Package's version(str) if package was found, otherwise return None.

Args: dictionary (dict): all the values will be reduced Reduce the values in the dictionary from all processes so that all processes

This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, he

In some inplace operations, we may change the box_mode of Boxes, this mode keeper context will help us to keep the box_mode unchanged

create an efficientnet model according to name. Args: model_name (str): Name for efficientnet. weights_path (None or

Calculates time since `start` and formats as a string.

Implementation from https://github.com/kuangliu/torchcv/blob/master/torchcv/utils/box.py with slight modifications. Assume t

Calculate and round number of filters based on width multiplier. Use width_coefficient, depth_divisor and min_depth of global_params. Args

Performs a single optimization step.

:param visual_tokens: B, seq, h :param position_2d_embeddings: B, seq, h :param position_1d_embeddings: B, seq, h :p

Computes the area of all the boxes. Returns: torch.Tensor: a vector with areas of each box.