引用karpathy · 2014-11
“Edit: And seemingly also from Berkeley Long-term Recurrent Convolutional Networks for Visual Recognition and Description”
补充一篇同样来自 Berkeley 的相近工作,对照 Stanford 的图像描述文章
把 CNN 与 LSTM 结合,用于视频识别与图像、视频描述
论文探讨既深于空间、又深于时间的循环卷积模型,提出可端到端训练的循环卷积架构,并在视频识别、图像描述与检索、视频叙述等任务上验证。这类“长期 RNN”能把变长输入(如视频帧)映射为变长输出(如文字),并与卷积网络联合训练,作者称其优于各部分分别定义或优化的最先进模型。
评论者把它当作用 CNN 加 RNN 处理视觉任务的例子,如有人拿它回应图像描述报道,也有人用它说明卷积 LSTM 的准确率。
“Edit: And seemingly also from Berkeley Long-term Recurrent Convolutional Networks for Visual Recognition and Description”
“Yes, yes it can: Long-term Recurrent Convolutional Networks for Visual Recognition and Description (Long-term Recurrent Convolutional Networks for Visual Recognition and Description)”
“You can get 89% accuracy with a convolutional LSTM Long-term Recurrent Convolutional Networks for Visual Recognition and Description”