def predict_behavior(self, image, bboxes, crop_height=112, crop_width=112):
width, height = image.size
boxes_data = []
for bbox in bboxes:
x1, y1, x2, y2 = bbox[:4]
boxes_data.append(
[y1 / height, x1 / width, y2 / height, x2 / width])
image = np.array(image, dtype=np.float32)
image = torch.from_numpy(image.transpose((2, 0, 1)))
image_data = image.unsqueeze(0)
image_data = torch.FloatTensor(image_data)
boxes_data = torch.FloatTensor(boxes_data)
boxes_index_data = torch.IntTensor([0 for _ in range(len(bboxes))])
image_torch = to_varabile(image_data) # N * C * H * W
boxes = to_varabile(boxes_data)
box_index = to_varabile(boxes_index_data)
crops_torch = CropAndResizeFunction(crop_height, crop_width,
0)(image_torch, boxes, box_index)
crops_torch = crops_torch.sub_(self.mean).div_(self.std)
outputs = self.model(crops_torch)
score = self.softmax(outputs)
pred = score.cpu().data.numpy()
return pred
def predict_behavior(self,
image_torch,
boxes,
box_index,
crop_height=112,
crop_width=112):
"""
predict batch images
"""
crops_torch = CropAndResizeFunction(crop_height, crop_width,
0)(image_torch, boxes, box_index)
crops_torch = crops_torch.sub_(self.mean).div_(self.std)
outputs = self.model(crops_torch)
score = self.softmax(outputs)
pred = score.cpu().data.numpy()
return pred