search

WongKinYiu / ScaledYOLOv4

Scaled-YOLOv4: Scaling Cross Stage Partial Network
GNU General Public License v3.0
2.02k stars 570 forks source link

ScaledYolov4 in Production and Scalability Aspects #168

Open shardulparab97 opened 3 years ago

shardulparab97 commented 3 years ago

I have successfully used and integrated scaled yolo V4 into my object detection pipeline. A big thanks to @WongKinYiu and the authors of Scaled Yolo v4 for this amazing repository and model.

Currently, I want to take my project forward into production, with a specific focus on scalability Would like to ask about any first-hand experiences of taking this specific model(I am using a custom trained P6 version) to production. Would love to know about go-to plans for production (in general too) and also any specific challenges or pointers for this specific model for production.

Additionally, the P6 custom trained model is highly accurate for my use case and now the focus is majorly on scalability and speed.

cyrilzakka commented 3 years ago

Did you happen to deploy it on mobile devices? If so, would you mind sharing your experience or some code?

shardulparab97 commented 3 years ago

@cyrilzakka - No not yet, just starting to plan about production and current focus for me is on GPU systems such as P40, T4, etc. In case I do take it up in the future, I'll definitely post on this issue thread :)

EconML commented 3 years ago

@shardulparab97 this is a bit off topic, but when you say 'successfully used and integrated' does this include model export after training?

xjinai commented 3 years ago

@shardulparab97 This is excellent news! Did you try to train Scaled YOLOv4 using multi-GPU?