praeclarumjj3/BackBone-Profile

Inference Time Performance stats for various backbone networks.

ResNet Profile

Contents

1. Overview
2. Setup Instructions
3. Repository Overview
4. Reproduction
5. Results

1. Overview

This repo contains the folowing Performance Stats for a few popularly used backbone networks in the field of Computer Vision:

- Inference Time on a GTX 2080Ti

- Inference Time on a TitanXP

- Infernce Time on a CPU

- Memory Report during Inference

- Model Structures

I have performed experiments on two types of inputs:

Size Format: (B,C,H,W)

• Cityscapes: Input of size = (1,3,1024,2048)

• PASCAL-VOC-2012: Input of size = (1,3,500,334)

2. Setup Instructions

You can setup the repo by running the following commands:

$ git clone https://github.com/praeclarumjj3/BackBone-Profile.git

$ pip install -r requirements.txt

3. Repository Overview

The repository contains the following architecture:

• MobileNetV2 Profiler - Scripts and stats for inference performance of MobileNet-V2.

• ResNet Profiler - Scripts and stats for inference performance of various variants of ResNet.

• Xception Profiler - Scripts and stats for inference performance of Xception.

4. Reproduction

• Refer to the README.md of the corresponding architectures.

5. Results

All the experiments are performed with a batch size=1 and 300 iterations.

Performance on Cityscapes

Model	Inference Time (ms) [2080Ti]	Inference Time (ms) [TitanXP]	FPS [2080Ti]	FPS [TitanXP]	Allocated Memory (MB)	# Params (M)
ResNet-18	18.719	23.622	53.42	42.33	68.69	11.689
ResNet-34	31.779	38.588	31.46	25.91	108.16	21.797
ResNet-50	61.397	82.334	16.28	12.14	121.73	25.557
ResNet-101	100.426	122.491	9.95	8.16	194.65	44.549
MobileNet-V2	33.627	54.314	29.73	18.41	37.58	3.504
Xception	77.079	144.919	12.97	6.90	111.45	22.855

Performance on PASCAL-VOC-2012

Model	Inference Time (ms) [2080Ti]	Inference Time (ms) [TitanXP]	FPS [2080Ti]	FPS [TitanXP]	Allocated Memory (MB)	# Params (M)
ResNet-18	2.547	2.940	392.61	340.13	46.60	11.689
ResNet-34	5.197	4.959	192.41	201.65	85.20	21.797
ResNet-50	7.628	8.927	131.09	112.01	100.23	25.557
ResNet-101	12.579	14.509	79.49	68.92	172.65	44.549
MobileNet-V2	5.570	5.795	179.53	172.56	14.49	3.504
Xception	7.919	12.042	126.27	83.04	89.36	22.855

Inference Time (CPU)

Performance on Cityscapes

Model	Inference Time (ms)	FPS
ResNet-18	566.75	1.76
ResNet-34	807.57	1.23
ResNet-50	1626.05	0.61
ResNet-101	2344.98	0.42
MobileNet-V2	560.022	1.78
Xception	2782.874	0.35

Performance on PASCAL-VOC-2012

Model	Inference Time (ms)	FPS
ResNet-18	55.79	17.92
ResNet-34	78.77	12.69
ResNet-50	133.71	7.47
ResNet-101	223.59	4.47
MobileNet-V2	70.180	14.24
Xception	229.000	4.36