Performance drops seriously when do int8 calibration

Tiruna · July 23, 2019, 9:02am

I use TensorRT-5.0 for int8 calibration. The origin model is an object detection caffemodel.
I find the performance drops sharply when the gorundtruth objects are small in image.
I crop the origin image to make the object bigger, but it just makes little sense. It will be worser if the gorundtruth objects include both small and big objects. Many objects are lost in inferring using int8 calibration model.

I would like to know how dose TensorRT do the int8 calibration. Why it failed in small objects?

k.sambasivarao13 · July 23, 2019, 9:27am

Hi, did you follow ssd sample? If so, what modifications you made in prototxt?

Tiruna · July 24, 2019, 2:23am

I did not use ssd sample. I just train a detection model and use tensorrt to do int8 calibration.

Topic		Replies	Views
convert a detect model to int8, the performance drops a lot TensorRT	1	1007	March 26, 2019
Accuracy of TensorRT in INT8 Deep Learning (Training & Inference) tensorrt	0	343	June 17, 2020
Calibration file problem between different versions of tensorrt? TensorRT	5	562	July 18, 2023
Int8 calibration on tensorrt 5 is much slower than tensorrt 4 TensorRT	1	815	February 28, 2019
Differences in performance between onnx models in Pytorch and TensorRT TensorRT	1	2227	June 17, 2019
TensorRT INT8 inference, the result is totally wrong! TensorRT	7	837	May 13, 2020
tensorRT int8 GPU-Accelerated Libraries	0	901	June 8, 2017
Tensorrt quantization bad performance TensorRT	3	699	November 20, 2020
INT8 Inference with low AP TensorRT	1	475	November 18, 2021
Int8 callibration with point cloud TensorRT	1	409	February 27, 2023

Performance drops seriously when do int8 calibration

Related topics