Create a SegLink detection classifier on a feature layer

Create or reuse a variable ARGS name: variable name shape: variable shape init_method: 'zero', 'kaiming', 'xavier', or (mean, std)

Convert image data from tensorflow to displayable format

Positive classification loss and hard negative classificatin loss ARGS scores: [n, n_classes] status: int [n] node or link matchi

Read image and groundtruth list. RETURN `image_paths`: list of image file paths `gt_paths`: list of groundtruth file paths

Construct the main body of a SegLink model

Visualize rotated bounding boxes ARGS ax: pyplot axis rboxes: array [n_rboxes, 5] or [5], (cx, cy, w, h, theta) colors: rboxes colors

Make a protobuf example. ARGS `image_binaries`: str, image jpeg binaries `annot_dict`: dict, annotations RETURN `exampl

Parse groundtruth annotations. ARGS `annot_file_path`: annotation file path RETURN `annot_dict`: dictionary of groundtruth an

Visualize encoded groundtruth ARGS ax: pyplot axis match_status: int [map_h, map_w] match status local_gt: [map_h, map_w, r

Visualize nodes and links of one example. ARGS `node_status`: int [map_h, map_w] `link_status`: int [map_h, map_w, n_links]

Load TensorFlow operator library.

Visualize links (local and cross) with colored lines ARGS `ax`: pyplot axis `link_status`: int [map_h, map_w, n_links] `image_size`:

* @brief Batch combine local rboxes. */

* @brief Clip rboxes by rectangular crop_bbox * @param rboxes, tensor [n_rboxes, 5] * @param crop_bbox, tensor [4] * @param clipped_rboxes, t

* @brief Combine local rboxes in one example. */

* @brief Encode groundtruth rboxes to local groundtruths in a batch * @param match_status, int tensor [batch, map_h, map_w] * @param local_pred,

* @brief Decode an example * @param match_status, int tensor data [map_h, map_w] * @param local_pred, tensor data [map_h, map_w, rbox_dim_] *

* @brief Decode segments of an example and find their group indices * @param all_node_status, node status of all detection layers * @param all_l

* @brief Encode groundtruth rboxes to local groundtruths in a batch * @param gt_rboxes, tensor [batch, n_gt_max, rbox_dim] * @param gt_counts, i

* @brief Encode groundtruth rboxes to local gt * @param gt_rboxes, tensor data [n_gt_max, gt_dim] * @param gt_count, int, number of groundtruths

* @brief Decode a local prediction * @param local_pred_p, tensor data [rbox_dim_], (cx, cy, w, h, theta) * @param grid_cx, grid_cy, float * @

* @brief Encode groundtruth into regression targets of a local region * @param gt_rboxes, tensor data [n_gt_max, rbox_dim_] * @param gt_count, i

* @brief Convert polygons to rboxes * @param polygons, tensor [n_polygons, 8] * @param rboxes, tensor [n_polygons, 5] */

* @brief Convert polygons to proj_polygons * @param polygons, tensor [n_polygons, 8] * @param crop_bbox, tensor [4] * @param proj_polygons, t