YOLT论文笔记+Transform_DOTA论文笔记

又开始读YOLT论文，本来是想直接实践的，但由于数据集过大，而且也确实不了解整个流程，所以重新看论文，把重要的几个工作点记录下来。

3 YOU ONLY LOOK TWICE
网络设计部分+预处理后处理
3.1 网络结构
this new architecture is optimized for small, densely packed objects
3.2 测试过程
At test time,we partition testing images of arbitrary size into manageable cutouts and run each cutout through our trained model
3.3 后处理过程
The final step in the object detection pipeline therefore seeks to stitch together the hundreds or thousands of testing chips into one final image strip.
4 TRAINING DATA
Training data is collected from small chips of large images from three sources: Digital Globe satellites,Planet satellites,and aerial platforms.
6.OBJECT DETECTION RESULTS
6.1 通用分类器目标检测结果
6.2 消除尺度难题
6.3 两个分类器结果

Transform_DOTA论文：
4 Experiments and Analysis
4.1 Datasets
—DOTA：简单介绍一下DOTA数据集并且说明了DOTA数据集处理的方法
首先要关注的就是数据集处理部分。
数据增强：
1）resize：we resize the image at two scales(1.0 and 0.5) for training and testing
2）crop： After image rescaling, we crop a series of 1024×1024 patches from the original images with a stride of 824
3）rotation：For those categories with a small number of samples, we do a rotation augmentation randomly from 4 angles (0,90,180,270) to simply avoid the effect of an imbalance between different categories
通过两篇论文，肯定是需要先做一个crop操作，因为太大了。
—HRSC2016：
数据增强：
flip，resize to (512,800)