How to use the openpifpaf.transforms.EVAL_TRANSFORM function in openpifpaf

def single_image(self, b64image):
        imgstr = re.search(r'base64,(.*)', b64image).group(1)
        image_bytes = io.BytesIO(base64.b64decode(imgstr))
        im = PIL.Image.open(image_bytes).convert('RGB')

        target_wh = self.width_height
        if (im.size[0] > im.size[1]) != (target_wh[0] > target_wh[1]):
            target_wh = (target_wh[1], target_wh[0])
        if im.size[0] != target_wh[0] or im.size[1] != target_wh[1]:
            print('!!! have to resize image to', target_wh, ' from ', im.size)
            im = im.resize(target_wh, PIL.Image.BICUBIC)
        width_height = im.size

        start = time.time()
        preprocess = openpifpaf.transforms.EVAL_TRANSFORM
        processed_image_cpu, _, __ = preprocess(im, [], None)
        processed_image = processed_image_cpu.contiguous().to(self.device, non_blocking=True)
        print('preprocessing time', time.time() - start)

        all_fields = self.processor.fields(torch.unsqueeze(processed_image.float(), 0))[0]
        keypoint_sets, scores = self.processor.keypoint_sets(all_fields)

        # normalize scale
        keypoint_sets[:, :, 0] /= processed_image_cpu.shape[2]
        keypoint_sets[:, :, 1] /= processed_image_cpu.shape[1]

        return keypoint_sets, scores, width_height