Training with a single RTX 3090

hustvl / MapTR

[ICLR'23 Spotlight & IJCV'24] MapTR: Structured Modeling and Learning for Online Vectorized HD Map Construction

MIT License

1.14k stars 174 forks source link

Training with a single RTX 3090 #69

Open lrx02 opened 1 year ago

lrx02 commented 1 year ago

I trained the method on a single RTX 3090 card and could not achieve results in the paper. I have set args.autoscale_l = True, can you provide parameters for single GPU? The following results are after 14 epochs. divider: 0.41260679562886554 ped_crossing: 0.3006982629497846 boundary: 0.4218177944421768 map: 0.3783742843402756

LegendBC commented 1 year ago

Sorry, we have not tested the single card setting. You can refer to https://github.com/facebookresearch/detr/issues/48#issuecomment-638689380 for scaling learning rate in your single-card setting.

dynamic721 commented 4 months ago

@lrx02 Hello! How did you solve this problem in the end?