Skip to content

[Attention] 超级视客营 MMRazor 🚀🚀🚀 #353

Description

@HIT-cwh

活动介绍

大家好,第一期 OpenMMLab 超级视客营实训活动开始啦!超级视客营实训活动提供十七个方向、上百个不同难度的任务供大家选择,不管你是初涉 AI 的新手还是资深炼丹师,都有适合你的任务供你选择。助力大家上手 OpenMMLab 开源算法库并参与项目建设。本期活动联合北京超级云计算中心,提供算力支持,为大家开发保驾护航。

活动参与方式:选择你感兴趣的任务,在 OpenMMLab 官网提交报名表。完成匹配后,即可和导师对接制定任务规划,开始上手开发。根据不同任务要求在对应的地址提交代码结果,出题方初步 review 通过后即可领取下一个任务或者坐等领奖。活动详情戳:OpenMMLab 官网活动页

任务列表

任务题目 任务描述 技术标签 难易程度 积分数
测试 MMRazor 通道依赖解析工具在 MMPOSE 上的鲁棒性 MMRazor拥有一套对模型通道依赖进行自动化解析的工具,以对模型进行剪。现需要在OpenMMLab各个repo上测试该项功能的鲁棒性,并进行简单的适配, 适配指南 熟悉 python,PyTorch 基础任务 20
测试 MMRazor 通道依赖解析工具在 MMRotate 上的鲁棒性 MMRazor拥有一套对模型通道依赖进行自动化解析的工具,以对模型进行剪。现需要在OpenMMLab各个repo上测试该项功能的鲁棒性,并进行简单的适配。 适配指南 熟悉 python,PyTorch 基础任务 20
测试 MMRazor 通道依赖解析工具在 MMFlow 上的鲁棒性 MMRazor拥有一套对模型通道依赖进行自动化解析的工具,以对模型进行剪。现需要在OpenMMLab各个repo上测试该项功能的鲁棒性,并进行简单的适配。适配指南 熟悉 python,PyTorch 基础任务 20
MovingAverageMinMaxObserver 参照pytorch中的MovingAverageMinMaxObserver,进行mmrazor的功能适配 熟悉 python,PyTorch 基础任务 20
FixedQParamsObserver 参照pytorch中的FixedQParamsObserver,进行mmrazor的功能适配 熟悉 python,PyTorch 基础任务 20
FixedQParamsFakeQuantize 参照 pytorch 中的 FixedQParamsFakeQuantize,进行 mmrazor 的功能适配 熟悉 python,PyTorch 基础任务 20
FusedMovingAvgObsFakeQuantize 参照 pytorch 中的 FusedMovingAvgObsFakeQuantize, 进行 mmrazor 的功能适配 熟悉 python,PyTorch 基础任务 20
AT ( cifar 数据集上的 kd 算法) 复现分类蒸馏算法:Paying More Attention to Attention: Improving the Performance of Convolutional Neural Networks via Attention Transfer
可参考源码:https://github.com/HobbitLong/RepDistiller
目标:在 cifar 数据集上复现精度
熟悉 python,PyTorch 中等任务 50
SP ( cifar 数据集上的 kd 算法) 复现分类蒸馏算法:Similarity-Preserving Knowledge Distillation
参考源码:https://github.com/HobbitLong/RepDistiller
目标:在 cifar 数据集上复现精度
熟悉 python,PyTorch 中等任务 50
CC ( cifar 数据集上的 kd 算法) 复现分类蒸馏算法: Correlation Congruence for Knowledge Distillation
可参考源码:https://github.com/HobbitLong/RepDistiller
目标:在 cifar 数据集上复现精度
熟悉 python,PyTorch 中等任务 50
VID ( cifar 数据集上的 kd 算法) 复现分类蒸馏算法:Variational Information Distillation for Knowledge Transfer
可参考源码:https://github.com/HobbitLong/RepDistiller
目标:在 cifar 数据集上复现精度
熟悉 python,PyTorch 中等任务 50
PKT ( cifar 数据集上的 kd 算法) 复现分类蒸馏算法:Probabilistic Knowledge Transfer for deep representation learning
可参考源码:https://github.com/HobbitLong/RepDistiller
目标:在 cifar 数据集上复现精度
熟悉 python,PyTorch 中等任务 50
AB ( cifar 数据集上的 kd 算法) 复现分类蒸馏算法:Knowledge Transfer via Distillation of Activation Boundaries Formed by Hidden Neurons
可参考源码:https://github.com/HobbitLong/RepDistiller
目标:在 cifar 数据集上复现精度
熟悉 python,PyTorch 中等任务 50
FT ( cifar 数据集上的 kd 算法) 复现分类蒸馏算法:Paraphrasing Complex Network: Network Compression via Factor Transfer
可参考源码:https://github.com/HobbitLong/RepDistiller
目标:在 cifar 数据集上复现精度
熟悉 python,PyTorch 中等任务 50
FSP ( cifar 数据集上的 kd 算法) 复现分类蒸馏算法:A Gift from Knowledge Distillation: Fast Optimization, Network Minimization and Transfer Learning
可参考源码:https://github.com/HobbitLong/RepDistiller
目标:在 cifar 数据集上复现精度
熟悉 python,PyTorch 中等任务 50
NST ( cifar 数据集上的 kd 算法) 复现分类蒸馏算法:Like what you like: knowledge distill via neuron selectivity transfer
可参考源码:https://github.com/HobbitLong/RepDistiller
目标:在 cifar 数据集上复现精度
熟悉 python,PyTorch 中等任务 50

活动报名地址:报名表地址

根据任务难度可以获得对应积分,兑换不同奖品。另外完成任务后在知识社区发布学习心得即可获得额外积分(记得主动找小助手领取哦)。

活动交流群:群二维码

有任何疑问欢迎大家加入群聊或者 Issue 下参与讨论,快来完成挑战,加入 OpenMMLab 贡献者队伍吧~

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions