The h5 of mask-rcnn is successfully converted to uff, but the result of uff is very different from h5

Description

according to sampleUffMaskRCNN/readme.md, I successfully converted h5 to uff, and the executable files of sampleUffMask and trtexec can load the uff file correctly。
Then I use opencv to load my own test image, the code is as follows:

//The following code replaces the func of readPPMFile
auto img_bgr = cv::imread(filename);
cv::Mat img_rgb = img_bgr;
cv::cvtColor(img_bgr, img_rgb, cv::COLOR_BGR2RGB);
    
ppm.fileName = filename;
ppm.magic = "P6";
ppm.max = 255;
ppm.h = img_rgb.rows;
ppm.w = img_bgr.cols;
ppm.buffer.assign(img_rgb.data, img_rgb.data + img_rgb.total() * img_rgb.channels());

But i didn’t get any output…, I checked the original h5 with the same image and got it right
I also used mask_rcnn_coco.h5 mentioned in readme.md, the output of the converted uff is consistent with the output mentioned in readme.md

My h5 model is trained on my own dataset based on the nucleus.py , and has not changed other than the dataset and marcnn/config.py

Is there any way for me to determine where the problem is or is there any special settings I need to operate?

Environment

TensorRT Version: TensorRT-7.2.1.6.Ubuntu-18.04.x86_64-gnu.cuda-10.2.cudnn8.0.tar.gz
GPU Type: rtx 2080ti
Nvidia Driver Version: NVRM version: NVIDIA UNIX x86_64 Kernel Module 455.38 Thu Oct 22 06:06:59 UTC 2020
CUDA Version: 10.2 with patch 10.2.1 and 10.2.2
CUDNN Version: 8.0.4.30
Operating System + Version: ubuntu 18.04
Baremetal or Container (if container which image + tag): nvcr.io/nvidia/tensorflow:19.10-py3

Hi, UFF and Caffe Parser have been deprecated from TensorRT 7 onwards, hence request you to try ONNX parser.

Please check the below link for the same.

Thanks!