Creates the matcher Params: cost_class: This is the relative weight of the classification error in the matching cost

return img, img_name(not abs_path), dir_name(not abs_path) e.g. img_name: []

rois [batch_size, num_anchor, 5[batch_idx. decoded_proposal]] gt_boxes [batch_size, num_anchor, 5[loc, labels]] use for proposal_target_l

Convert prior_boxes to (cx, cy, w, h) representation for comparison to center-size form ground truth data. Args: boxes: (tensor) poin

gt_bboxes: n * [x0, y0, x1, y1] transformed_anchors: [c_x, cy, w, h]

img: arr bgr h,w,3 bbox_labels: arr [[xmin, ymin, xmax, ymax, label]] return: sampled_bbox (list), img, bbox_labels, for img and bbox_lab

loc: torch.Tensor anchors: 2-d, torch.Tensor (cx, cy, w, h) boxes: 2-d, torch.Tensor (x0, y0, x1, y1)

Decode locations from predictions using priors to undo the encoding we did for offset regression at train time. Args: loc (tensor): lo

align

Draw one bounding box on image. Args: image (PIL.Image): a PIL Image object. bbox (np.array|list|tuple): (xmin, ymin, xmax, y

Draw bounding boxes on image. Args: img (np.ndarray or string): input image or image path. bboxes (np.array): bounding b

anchors: np.array (cx, cy, w, h) anchor_matched_gt np.array (x0, y0, x1, y1) ret: np.array (delta(cx) / anchors_w, delta(cy) / anchors_h,

anchors: Tensor (cx, cy, w, h) anchor_matched_gt Tensor (x0, y0, x1, y1) ret: Tensor (delta(cx) / anchors_w, delta(cy) / anchors_h, log(g

Performs the matching Params: outputs: This is a dict that contains at least these entries: "pred_logits": Tens

Generalized IoU from https://giou.stanford.edu/ The boxes should be in [x0, y0, x1, y1] format Returns a [N, M] pairwise matrix, where

gt_bbox: [x0, x1, y0, y1] transformed_anchors: [c_x, cy, w, h]

gt_bbox: [x0, x1, y0, y1] transformed_anchors: [c_x, cy, w, h]

gt_bboxes: n * [x0, x1, y0, y1] transformed_anchors: [c_x, cy, w, h]

Utility function for computing log_sum_exp while determining This will be used to determine unaveraged confidence loss across all examples in