Create tf records for the VGG SynthText dataset ARGS save_path: path to save the TF record data_root: the root folder for the datasets

Input data stream ARGS `record_path`: tf records file path RETURN `streams`: data streams

Log all variables defined in FLAGS and their values.

Log git version number and uncommitted changes if any.

Calculate the bounding boxes of rboxes ARGS `rboxes`: [n, 5] RETURN `bboxes`: [n, 4]

Convert rboxes to polygons ARGS `rboxes`: [n, 5] RETURN `polygons`: [n, 8]

Classification loss ARGS gt_labels: int32 [n] match_scores: [n, n_classes] RETURN loss

Setup a logger that simultaneously output to a file and stdout ARGS log_file_path: string, path to the logging file

Smooth L1 loss between offsets and encoded_gt ARGS offsets: [m?, 5], predicted offsets for one example gt_offsets: [m?, 5], correponding

Start Tensorboard at port 8000.

Preprocess test images and groundtruths ARGS streams: input data streams RETURN pstreams: preprocessed data streams

Preprocess training images and groundtruths ARGS streams: input data streams RETURN pstreams: preprocessed data streams

Upload experiment logs and events to remote server.

Visualize bounding boxes on images ARGS images: uint8 bboxes: [N,n,4], or [n,4], or list of [?,4], normalized bboxes scores: [N,n,n_c

Visualize joined rboxes.

Visualize local rboxes of each layer on separated axis. ARGS: `sess_outputs`: TF session outputs `save_dir`: saving directory

Visualize a grid of nodes with colored dots ARGS `ax`: pyplot axis `node_status`: int [map_h, map_w] `image_size`: int [2]

Visualize linked local rboxes on one figure.