Clip boxes to image boundaries.

Clip boxes to image boundaries.

Convert an image and RoIs within that image into network inputs.

Convert an image and RoIs within that image into network inputs.

Get features for all ann_ids in an image

Return - labels : Variable float (N, num_atts) - select_ixs: Variable long (n, )

For a given ann_id, we return - st_ann_ids: same-type neighbouring ann_ids (not include itself) - dt_ann_ids: different-type

Arguments: head_feats_dir: cache/feats/dataset_splitBy/net_imdb_tag, containing all image conv_net feats args: imdb_name,

Converts an image into a network input. Arguments: im (ndarray): a color image in BGR order Returns: blob (ndarray): a data blob h

Converts an image into a network input. Arguments: im (ndarray): a color image in BGR order Returns: blob (ndarray): a data blob h

Arguments: net_conv (Variable) : (1, 1024, H, W) im_info (float32) : [[ih, iw, scale]] ori_boxes (float32) : (n, 4) [x1y1x

labels: int32 (n, label_length) zeros padded in end return: list of sents in string format

Return - cxt_feats : ndarray (#ann_ids, topK, fc7_dim) - cxt_lfeats: ndarray (#ann_ids, topK, 5) - cxt_ann_ids: [[ann

returns -pool5 (n, 1024, 7, 7) -fc7 (n, 2048, 7, 7)

Returns head: float32 (1, 1024, H, W) im_info: float32 [[im_h, im_w, im_scale]]

Convert [x y w h] box format to [x1 y1 x2 y2] format.

Arguments: net_conv (Variable) : (1, 1024, H, W) im_info (float32) : [[ih, iw, scale]] ori_boxes (float32) : (n, 4) [x1y1x

Arguments: net_conv (Variable) : (1, 1024, H, W) im_info (float32) : [[ih, iw, scale]] ori_boxes (float32) : (n, 4) [x1y1x

Arguments: net_conv (Variable) : (1, 1024, H, W) im_info (float32) : [[ih, iw, scale]] ori_boxes (float32) : (n, 4) [x1y1x

Arguments: net_conv (Variable) : (1, 1024, H, W) im_info (float32) : [[ih, iw, scale]] ori_boxes (float32) : (n, 4) [x1y1x

Arguments: - img_path: img_file - boxes : ndaray [[xyxy]] (n, 4) in original image - labels : ndarray (n, ) Return: - mask

Arguments: - img_path: img_file - boxes : ndaray [[xyxy]] (n, 4) in original image - labels : ndarray (n, ) Return: - mask

- scores: Variable (cuda) (n, num_atts) after sigmoid range [0, 1] - labels:list of [[att, sc], [att, sc], ...

Input: sent_str_list: list of n sents in string format return int32 (n, label_length) zeros padded in end

return: int32 (num_sents, label_length) and picked_sent_ids

Sample #num_sents sents for each ref_id.

attribute_feats.shape = (sent_num, ann_num, 512), total_ann_score.shape = (sent_num, ann_num)

sub_phrase_emb, loc_phrase_emb, rel_phrase_emb.shape = (sent_num, 512), labels.shape = (sent_num, sent_length)

(sub_phrase_emb, sub_phrase_recons, loc_phrase_emb, loc_phrase_recons, rel_phrase_emb, rel_phrase_recons).shape=(sent_num, 512) weigh

Inputs: - input_labels: Variable long (batch, seq_len) Outputs: - output : Variable float (batch, max_len, hidden_size * num_dirs)

Arguments: net_conv (Variable): (1, 1024, H, W) im_info (float) : [[ih, iw, scale]] Returns: scores (ndarray): (num_rois, 8

Arguments: net_conv (Variable): (1, 1024, H, W) im_info (float) : [[ih, iw, scale]] Returns: scores (ndarray): (num_rois, 8

Only used for testing. Testing the above box_to_fc7 [passed]

Only used for testing. Testing the above box_to_fc7 [passed]

Convert [x1 y1 x2 y2] box format to [x y w h] format.