alibaba / alibaba/esod

关于 vanilla 与 ESOD 版本推理速度差异的疑问

Open
#21 2 comments 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
140
Forks
21
PR merge metrics
No merged PRs in 30d

Description

您好,

我在相同环境下分别训练并测试了 YOLOv5 和 RetinaNet 的 ESOD 版本 与 vanilla 版本,结果发现无论是哪种模型,vanilla 版本的推理速度都显著快于 ESOD 版本。

在输入图像尺寸为 1536×1536 的情况下,测试得到的 FPS 如下:

Image

我训练模型使用的命令为DATASET=visdrone MODEL=yolov5m(或retinanet) GPUS=0 BATCH_SIZE=8 IMAGE_SIZE=1536 EPOCHS=50 bash ./scripts/train.sh,并除了将train.sh中的model_type进行esod/vanilla的调整以外没有对代码部分进行任何修改。

我测试fps使用的命令为python test.py --data data/visdrone.yaml --weights runs/train/expn/weights/best.pt --batch-size 1 --img-size 1536 --device 0 --task measure。

请问是我的训练、测试过程中有什么地方有错误吗?如有误的话该如何调整以获取和论文结果相近的推理速度结果呢。

Contributor guide

No contributing guide indexed for this repository

Research direction

Start by reviewing scripts/train.sh and test.py, then reproduce the reported YOLOv5 and RetinaNet ESOD versus vanilla benchmarks with the commands and 1536×1536, batch-size 1 settings given in the issue. Compare the actual environment and measurement configuration with the paper; done means identifying the discrepancy and documenting the adjustment needed to obtain comparable inference speeds.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.