[Paper note] Aggregated Residual Transformations for Deep Neural Networks

Model

  • Solution space of Inception architecture is a strict subspace of the solution space of a single large layer (e.g., 5×5) operating on a high-dimensional embedding.
  • The transformations to be aggregated are all of the same topology.
  • Equivalent forms:
    equivalent
  • This strategy exposes a new dimension, which we call “cardinality” (the size of the set of transformations), as an essential factor in addition to the dimensions of depth and width.

Experiemnt

  • See original paper for training details.
  • Increasing cardinality is more efficient than increasing depth/width.
  • ImageNet
    • 224x224: 20.4% top-1, 5.3% top-5
    • 320x320/crop-to-299x299: 19.1% top-1, 4.4 top-5
  • ImageNet 5K (5000 class)
  • CIFAR
    • -10 3.58%
    • -100 17.31%
  • COCO detection
    • 51.9 AP@0.5
    • 30.0 AP
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值