search

mercedes-benz / selfsupervised_flow

Code for Paper "Self-Supervised LiDAR Scene Flow and Motion Segmentation"
MIT License
57 stars 11 forks source link

Question about training details #20

Open LYFFF666 opened 11 months ago

LYFFF666 commented 11 months ago

Thank you for such a great work of open source source. When I was training, I encountered the problem of insufficient video memory, I would like to ask you what graphics card model you used during training? What is the memory size?

baurst commented 11 months ago

Hi, thanks for the interest in our paper. We used V100 with 16Gb RAM for our experiments (single GPU, batch size = 1).

LYFFF666 commented 11 months ago

Thank you very much for your reply and help, but when I use python unsup_flow/cli.py --prod -c default sota_us sota_net , I met this warning:

/root/autodl-tmp/selfsupervised_flow-master/.venv/lib/python3.6/site-packages/tensorflow/python/framework/indexed_slices.py:449: UserWarning: Converting sparse IndexedSlices(IndexedSlices(indices=Tensor("gradient_tape/our_pillar_model/head_decoder_forward/construct_static_aggregation_4/static_aggregated_flow/weighted_pc_alignment/RaggedTile_4/Reshape_3:0", shape=(None,), dtype=int64), values=Tensor("gradient_tape/our_pillar_model/head_decoder_forward/construct_static_aggregation_4/static_aggregated_flow/weighted_pc_alignment/RaggedTile_4/Reshape_2:0", shape=(None, 3), dtype=float32), dense_shape=Tensor("gradient_tape/our_pillar_model/head_decoder_forward/construct_static_aggregation_4/static_aggregated_flow/weighted_pc_alignment/RaggedTile_4/Cast:0", shape=(2,), dtype=int32))) to a dense Tensor of unknown shape. This may consume a large amount of memory. "shape. This may consume a large amount of memory." % value)

I wonder if this training instruction is correct, is there any configuration option in the config file that I didn't notice?