Apppublic

xdecoder/Instruct-X-Decoder

sourceHugging Faceafl-3.0updated 3y agoView on Hugging Face

163likes

roottasks__pycache__ref_cap.cpython-38.pyc

ref_cap.cpython-38.pyc35 linesDownload Raw Back to __pycache__

1U

2E$�c�3�@s�ddlZddlmmZddlZddlmZddl	m4Z5ddlmZddl
mZgZe�e6jdejd��e7�e�ZgZe�e8jdejd��e9�e�Ze�d	�Zd10d�ZdS)�N)�Image)�11transforms)�12Visualizer)�MetadataCatalog��)�
interpolationi�coco_2017_train_panopticc13Os |\}}t�����|}t|�}|jd}	|jd}14t�|�}|}t�|����ddd��	�}|�15d�rp|��n|dgg}
|d|
i|16|	d�g}|j�
|d�}|dddk��}dtj|d	d17dd�d��}d
ddg}t|td�}|j|����d||d�}|��}d|����dk�rHtj	��t�|�ddfW5QR�S|d}t|�}t�|�}t�|�}t�|����ddd��	�}|d|d�g}|�18d��r�|�dd�n|}|jjjj j!�"|�}t�#|�ddd�f}|jj$|d|id�}|dd}W5QRXtj	��t�|�|dfS)Nr���.�texts)�image�19groundings�height�width������grounding_mask)N)rr�nearest)�modeg�������?g�������?g000000�?)�metadata)�color�text�zn/a)r
�image_id�captioning_mask��token)�extra�captioning_text)%�torch�no_grad�
transform_grd�size�np�asarray�20from_numpy�copy�permute�cuda�endswith�strip�model�evaluate_grounding�float�F�interpolate�boolr�metedata�draw_binary_mask�cpu�numpy�	get_image�sum�empty_cacher�	fromarray�
transform_ret�replace�sem_seg_head�	predictor�lang_encoder�	tokenizer�encode�tensor�evaluate_captioning)r+r
r�inpainting_text�args�kwargs�21model_last�	model_cap�	image_orirrZ22image_ori_�imagesZtexts_input�batch_inputs�outputsZgrd_maskZ	grd_mask_r�visual�demo�resZ23token_textrr�rN�&/data/arXiv/demo/Demo/tasks/ref_cap.py�referring_captioningsD242526 2728293031rP)rZtorch.nn.functional�nn�32functionalr.r4r#�PILr�torchvisionr�utils.visualizerr�detectron2.datar�t�append�Resize�BICUBIC�Composer9r!�getr1rPrNrNrNrO�<module>s333435