Detected object in TensorRT VS Darknet model

AI & Data Science Deep Learning (Training & Inference) TensorRT
1

when I convert my YOLOv3 model to tensorRT, some objects that can be detected in YOLOv3 are not detected in TensorRT with FP32 mode. The project is based on this repository:[url] https://github.com/lewes6369/TensorRT-Yolov3/[/url]

Objects detected in Darknet YOLOv3

1920×1208 132 KB

[/img]

Objects detected in TensorRT YOLOv3

1920×1208 139 KB

Should I modify the training data or modify some settings in TensorRT ?

2

what version of TRT are you using? and just to confirm, GitHub - lewes6369/TensorRT-Yolov3: TensorRT for Yolov3 is the repro that demonstrates the issue are you are seeing?

3

HI NVES, the version of TensorRT I’m using is TensorRT 5.0.2.6. The example YOLOv3 model in the repository works fine, but when I train the architecture with my own dataset, the problem appears.

4

Hello,

to help us debug, please share a repro package that contains the inference code, model, and dataset used that demonstrate the issues you are seeing. you can dm me a google drive, dropbox link, or github link.

5

HI NVES, it seems my preprocess function cutoff some area from the original image. Now this issue is solved, thank you.