def forward(self, img_id, img_shape, inputs):
outputs = []
boxes = []
scores = []
downsample = 32
feats = self.backbone(inputs)
route = None
for idx, feat in enumerate(feats[::-1]):
if idx > 0:
feat = paddle.concat(x=[route, feat], axis=1)
route, tip = self.yolo_blocks[idx](feat)
block_out = self.block_outputs[idx](tip)
outputs.append(block_out)
if idx < 2:
route = self.route_blocks[idx](route)
route = F.resize_nearest(route, scale=2)
if self.model_mode != 'train':
anchor_mask = self.anchor_masks[idx]
mask_anchors = []
for m in anchor_mask:
mask_anchors.append(self.anchors[2 * m])
mask_anchors.append(self.anchors[2 * m + 1])
b, s = F.yolo_box(x=block_out,
img_size=img_shape,
anchors=mask_anchors,
class_num=self.num_classes,
conf_thresh=self.valid_thresh,
downsample_ratio=downsample)
boxes.append(b)
scores.append(paddle.transpose(s, perm=[0, 2, 1]))
downsample //= 2
if self.model_mode == 'train':
return outputs
preds = [
img_id,
F.multiclass_nms(bboxes=paddle.concat(boxes, axis=1),
scores=paddle.concat(scores, axis=2),
score_threshold=self.valid_thresh,
nms_top_k=self.nms_topk,
keep_top_k=self.nms_posk,
nms_threshold=self.nms_thresh,
background_label=-1)
]
if self.model_mode == 'test':
return preds
# model_mode == "eval"
return outputs + preds
def predict(self,
imgpath: str,
filelist: str,
visualization: bool = True,
save_path: str = 'result'):
'''
Detect images
Args:
imgpath(str): Image path .
filelist(str): Path to get label name.
visualization(bool): Whether to save result image.
save_path(str) : Path to save detected images.
Returns:
boxes(np.ndarray): Predict box information.
scores(np.ndarray): Predict score.
labels(np.ndarray): Predict labels.
'''
self.eval()
boxes = []
scores = []
self.downsample = 32
im = self.transform(imgpath)
h, w, c = utils.img_shape(imgpath)
im_shape = paddle.to_tensor(np.array([[h, w]]).astype('int32'))
label_names = utils.get_label_infos(filelist)
img_data = paddle.to_tensor(np.array([im]).astype('float32'))
outputs = self(img_data)
for i, out in enumerate(outputs):
anchor_mask = self.anchor_masks[i]
mask_anchors = []
for m in anchor_mask:
mask_anchors.append((self.anchors[2 * m]))
mask_anchors.append(self.anchors[2 * m + 1])
box, score = F.yolo_box(x=out,
img_size=im_shape,
anchors=mask_anchors,
class_num=self.class_num,
conf_thresh=self.valid_thresh,
downsample_ratio=self.downsample,
name="yolo_box" + str(i))
boxes.append(box)
scores.append(paddle.transpose(score, perm=[0, 2, 1]))
self.downsample //= 2
yolo_boxes = paddle.concat(boxes, axis=1)
yolo_scores = paddle.concat(scores, axis=2)
pred = F.multiclass_nms(bboxes=yolo_boxes,
scores=yolo_scores,
score_threshold=self.valid_thresh,
nms_top_k=self.nms_topk,
keep_top_k=self.nms_posk,
nms_threshold=self.nms_thresh,
background_label=-1)
bboxes = pred.numpy()
labels = bboxes[:, 0].astype('int32')
scores = bboxes[:, 1].astype('float32')
boxes = bboxes[:, 2:].astype('float32')
if visualization:
if not os.path.exists(save_path):
os.mkdir(save_path)
utils.draw_boxes_on_image(imgpath, boxes, scores, labels,
label_names, 0.5, save_path)
return boxes, scores, labels