def __get_raw_bbox_info(self, img):
with torch.no_grad():
im_data = torch.FloatTensor(1).cuda()
im_info = torch.FloatTensor(1).cuda()
num_boxes = torch.LongTensor(1).cuda()
gt_boxes = torch.FloatTensor(1).cuda()
box_info = torch.FloatTensor(1).cuda()
im_blob, im_scales = self.__get_image_blob(img)
im_info_np = np.array(
[[im_blob.shape[1], im_blob.shape[2], im_scales[0]]],
dtype=np.float32)
im_data_pt = torch.from_numpy(im_blob)
im_data_pt = im_data_pt.permute(0, 3, 1, 2)
im_info_pt = torch.from_numpy(im_info_np)
im_data.resize_(im_data_pt.size()).copy_(im_data_pt)
im_info.resize_(im_info_pt.size()).copy_(im_info_pt)
gt_boxes.resize_(1, 1, 5).zero_()
num_boxes.resize_(1).zero_()
box_info.resize_(1, 1, 5).zero_()
# forward
rois, cls_prob, bbox_pred, \
rpn_loss_cls, rpn_loss_box, \
RCNN_loss_cls, RCNN_loss_bbox, \
rois_label, loss_list = self.hand_detector(im_data, im_info, gt_boxes, num_boxes, box_info)
scores = cls_prob.data
boxes = rois.data[:, :, 1:5]
# hand side info (left/right)
lr_vector = loss_list[2][0].detach()
lr = torch.sigmoid(lr_vector) > 0.5
lr = lr.squeeze(0).float()
box_deltas = bbox_pred.data
stds = [0.1, 0.1, 0.2, 0.2]
means = [0.0, 0.0, 0.0, 0.0]
box_deltas = box_deltas.view(-1, 4) * torch.FloatTensor(stds).cuda() \
+ torch.FloatTensor(means).cuda()
box_deltas = box_deltas.view(1, -1, 4 * len(self.classes))
pred_boxes = bbox_transform_inv(boxes, box_deltas, 1)
pred_boxes = clip_boxes(pred_boxes, im_info.data, 1)
pred_boxes /= im_scales[0]
scores = scores.squeeze()
pred_boxes = pred_boxes.squeeze()
return scores, pred_boxes, lr
def __get_raw_hand_bbox(self, img):
with torch.no_grad():
im_data = torch.FloatTensor(1).cuda()
im_info = torch.FloatTensor(1).cuda()
num_boxes = torch.LongTensor(1).cuda()
gt_boxes = torch.FloatTensor(1).cuda()
box_info = torch.FloatTensor(1).cuda()
im_blob, im_scales = self.__get_image_blob(img)
im_info_np = np.array(
[[im_blob.shape[1], im_blob.shape[2], im_scales[0]]],
dtype=np.float32)
im_data_pt = torch.from_numpy(im_blob)
im_data_pt = im_data_pt.permute(0, 3, 1, 2)
im_info_pt = torch.from_numpy(im_info_np)
im_data.resize_(im_data_pt.size()).copy_(im_data_pt)
im_info.resize_(im_info_pt.size()).copy_(im_info_pt)
gt_boxes.resize_(1, 1, 5).zero_()
num_boxes.resize_(1).zero_()
box_info.resize_(1, 1, 5).zero_()
# forward
rois, cls_prob, bbox_pred, \
rpn_loss_cls, rpn_loss_box, \
RCNN_loss_cls, RCNN_loss_bbox, \
rois_label, loss_list = self.hand_detector(im_data, im_info, gt_boxes, num_boxes, box_info)
scores = cls_prob.data
boxes = rois.data[:, :, 1:5]
# hand side info (left/right)
lr_vector = loss_list[2][0].detach()
lr = torch.sigmoid(lr_vector) > 0.5
lr = lr.squeeze(0).float()
box_deltas = bbox_pred.data
stds = [0.1, 0.1, 0.2, 0.2]
means = [0.0, 0.0, 0.0, 0.0]
box_deltas = box_deltas.view(-1, 4) * torch.FloatTensor(stds).cuda() \
+ torch.FloatTensor(means).cuda()
box_deltas = box_deltas.view(1, -1, 4 * len(self.classes))
pred_boxes = bbox_transform_inv(boxes, box_deltas, 1)
pred_boxes = clip_boxes(pred_boxes, im_info.data, 1)
pred_boxes /= im_scales[0]
scores = scores.squeeze()
pred_boxes = pred_boxes.squeeze()
thresh_hand = 0.5
j = 2
inds = torch.nonzero(scores[:, j] > thresh_hand,
as_tuple=False).view(-1)
if inds.numel() > 0:
cls_boxes = pred_boxes[inds][:, j * 4:(j + 1) * 4]
cls_scores = scores[:, j][inds]
_, order = torch.sort(cls_scores, 0, True)
lr = lr[inds][order]
cls_boxes = cls_boxes[order]
cls_scores = cls_scores[order]
keep = nms(cls_boxes, cls_scores,
cfgg.TEST.NMS) # cfgg.TEST.NMS : 0.3
cls_boxes = cls_boxes[keep]
lr_selected = lr[keep]
cls_boxes_np = cls_boxes.detach().cpu().numpy()
lr_np = lr_selected.detach().cpu().numpy()
return cls_boxes_np, lr_np[:, 0]
# return cls_boxes_np, lr_np
else:
return None, None