[深度学习论文笔记][Video Classification] Long-term Recurrent Convolutional Networks for Visual Recognition a
2016-11-17 11:06
746 查看
Donahue, Jeffrey, et al. ”Long-term recurrent convolutional networks for visual recognition and description.” Proceedings of the IEEE Conference on Computer Vision and Pattern
Recognition. 2015. (Citations: 418).
1 Motivation
All the previous papers used local motion cues to get extra accuracy (e.g. half a second or so). What if the temporal dependencies of interest are much much longer?
2 Recurrent Convolutional Architecture
See Fig. The CNN part is time-invariant and independent at each timestep. In this paper, we focus on three specif tasks, see Fig.
2.1 Video Classification
Sequential input, fixed output. We use late fusion to merge per-timestep predictions. The CNN part can use the ImageNet pre-trained model.
We also consider both RGB and flow inputs. Flow is computed and transformed into a “flow image” by centering x and y flow values around 128 and multiplying by a scalar such
that flow values fall between 0 and 255. A third channel for the flow image is created by calculating the flow magnitude.
2.2 Image Captioning
Fixed input, sequential output. At each timestep, both the image features and the previous word are provided as inputs to the LSTM.
2.3 Video Captioning
Sequential input, sequence output. This is the encoder-decoder style. Input is processed and the decoder outputs are ignored for the first T timesteps, and the predictions are made and “dummy” inputs are ignored for the latter T ′ timesteps.
Recognition. 2015. (Citations: 418).
1 Motivation
All the previous papers used local motion cues to get extra accuracy (e.g. half a second or so). What if the temporal dependencies of interest are much much longer?
2 Recurrent Convolutional Architecture
See Fig. The CNN part is time-invariant and independent at each timestep. In this paper, we focus on three specif tasks, see Fig.
2.1 Video Classification
Sequential input, fixed output. We use late fusion to merge per-timestep predictions. The CNN part can use the ImageNet pre-trained model.
We also consider both RGB and flow inputs. Flow is computed and transformed into a “flow image” by centering x and y flow values around 128 and multiplying by a scalar such
that flow values fall between 0 and 255. A third channel for the flow image is created by calculating the flow magnitude.
2.2 Image Captioning
Fixed input, sequential output. At each timestep, both the image features and the previous word are provided as inputs to the LSTM.
2.3 Video Captioning
Sequential input, sequence output. This is the encoder-decoder style. Input is processed and the decoder outputs are ignored for the first T timesteps, and the predictions are made and “dummy” inputs are ignored for the latter T ′ timesteps.
相关文章推荐
- 【论文阅读笔记】CVPR2015-Long-term Recurrent Convolutional Networks for Visual Recognition and Description
- 深度学习论文笔记:Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition
- 深度学习论文笔记-Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition
- [深度学习论文笔记][Semantic Segmentation] Recurrent Convolutional Neural Networks for Scene Labeling
- Long-term Recurrent Convolutional Networks for Visual Recognition and Description
- 《Long-term Recurrent Convolutional Networks for Visual Recognition and Description》论文翻译
- long term recurrent convolutional networks for visual recognition and description
- [深度学习论文笔记][Video Classification] Two-Stream Convolutional Networks for Action Recognition in Videos
- 深度学习笔记空间金字塔池化阅读笔记Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition
- 深度学习笔记(一)空间金字塔池化阅读笔记Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition
- [深度学习] Very Deep Convolutional Networks for Large-Scale Image Recognition(VGGNet)阅读笔记
- [深度学习论文笔记][Image Classification] Very Deep Convolutional Networks for Large-Scale Image Recognitio
- 【论文阅读笔记】Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition
- [深度学习论文笔记][Semantic Segmentation] Fully Convolutional Networks for Semantic Segmentation
- RCNN学习笔记(1):《Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition》论文笔记
- 论文笔记|Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition
- [深度学习论文笔记][Visualizing] Deep Inside Convolutional Networks Visualising Image Classification
- 【深度学习论文笔记:Recognition】:Deep Neural Networks for Object Detection
- 论文笔记之:Spatially Supervised Recurrent Convolutional Neural Networks for Visual Object Tracking