基于区块链的毕业设计Human Motion Prediction – 人体运动预测

本文提供基于区块链的毕业设计国外最新区块链项目源码下载,包括solidity,eth,fabric等blockchain区块链,基于区块链的毕业设计Human Motion Prediction – 人体运动预测 是一篇很好的国外资料

Human Motion Prediction

General Information

Authors: Pratyush Singh, Rohit Kaushik, Melvin Ott; Group: BruteForce

Abstract

Modelling human motion is a challenging task in computer vision and graphics since human body movement can change drastically from one environment to another. It’s an important task for developing and deploying autonomous agents. In order to tackle this issue, we take inspiration from the Structured Prediction Layer (SPL) which decomposes the pose into individual joints and can be augmented with different neural network architectures. We introduce the SPL dropout layer and describe its effect on the prediction scores. Using the per joint loss instead ofthe standard mean squared error as well as a residual connection for modelling velocities help us stay afloat on the top of leaderboard in the Machine Perception course project at ETH Zurich.

Recreating our best submission

You can use the following command:

bsub -n 6 -W 24:00 -R "rusage[mem=8192, ngpus_excl_p=1]" -R "select[gpu_model0==GeForceGTX1080Ti]" python train.py --data_dir /cluster/project/infk/hilliges/lectures/mp20/project4 --save_dir ./experiments   --model_type rnn_spl --spl_dropout --spl_dropout_rate 0.0 --input_hidden_size 256 --input_hidden_layers 1 --output_hidden_layers 1 --output_hidden_size 128 --input_dropout_rate 0.04 --num_epochs 700 --experiment_name rnnspl    --learning_rate_decay_rate 0.91 --residual_velocity --cell_size 2048 --cell_type blstm --early_stopping_tolerance 60

and evaluate with:

bsub -n 6 -W 4:00 -R "rusage[mem=1024, ngpus_excl_p=1]"  -R "select[gpu_model0==GeForceGTX1080Ti]" -o outvalidate.txt python evaluate_test.py --data_dir /cluster/project/infk/hilliges/lectures/mp20/project4 --save_dir ./experiments --model_id <model_id> --export

Notes

Make sure to not use a GTX 1080 on the cluster as this lead to problems in our experience

Proper licensing and permission required to use the codes.


人体运动预测

General Information

作者:Pratyush Singh,Rohit Kaushik,Melvin Ott;Group:BruteForce建模人体运动在计算机视觉和图形中是一项具有挑战性的任务,因为人体运动可以从一个环境急剧变化到另一个环境。开发和部署自治代理是一项重要的任务。为了解决这一问题,我们从结构化预测层(SPL)中得到启发,该层将姿势分解为单个关节,并可以使用不同的神经网络结构进行扩充。我们引入了SPL-dropout层,并描述了它对预测分数的影响。使用每个关节的损失而不是标准的均方误差,以及建模速度的剩余连接,帮助我们在苏黎世ETH机器感知课程项目中保持领先地位。

Abstract

您可以使用以下命令:

Recreating our best submission

并使用进行评估:

bsub -n 6 -W 24:00 -R "rusage[mem=8192, ngpus_excl_p=1]" -R "select[gpu_model0==GeForceGTX1080Ti]" python train.py --data_dir /cluster/project/infk/hilliges/lectures/mp20/project4 --save_dir ./experiments   --model_type rnn_spl --spl_dropout --spl_dropout_rate 0.0 --input_hidden_size 256 --input_hidden_layers 1 --output_hidden_layers 1 --output_hidden_size 128 --input_dropout_rate 0.04 --num_epochs 700 --experiment_name rnnspl    --learning_rate_decay_rate 0.91 --residual_velocity --cell_size 2048 --cell_type blstm --early_stopping_tolerance 60

请确保不要在群集上使用GTX 1080,因为这会导致我们的经验中出现问题

bsub -n 6 -W 4:00 -R "rusage[mem=1024, ngpus_excl_p=1]"  -R "select[gpu_model0==GeForceGTX1080Ti]" -o outvalidate.txt python evaluate_test.py --data_dir /cluster/project/infk/hilliges/lectures/mp20/project4 --save_dir ./experiments --model_id <model_id> --export

Notes

使用代码所需的正确许可和许可。

Proper licensing and permission required to use the codes.

部分转自网络,侵权联系删除区块链源码网

www.interchains.cc

https://www.interchains.cc/17373.html

区块链毕设网(www.interchains.cc)全网最靠谱的原创区块链毕设代做网站 部分资料来自网络,侵权联系删除! 最全最大的区块链源码站 !
区块链知识分享网, 以太坊dapp资源网, 区块链教程, fabric教程下载, 区块链书籍下载, 区块链资料下载, 区块链视频教程下载, 区块链基础教程, 区块链入门教程, 区块链资源 » 基于区块链的毕业设计Human Motion Prediction – 人体运动预测

提供最优质的资源集合

立即查看 了解详情