Focal Loss for Dense Object Detection
1.
prevents the vast number of easy negatives from overwhelming the detector during training.
在训练期间防止大量容易的负面因素使检测器不堪重负。
2.
Recent work on one-stage detectors, such asYOLO and SSD, demonstrates promising results, yielding faster detectors with accuracy within 10-40% relative to state-of-the-art two-stage methods.
最近关于单级探测器的工作，例如YOLO和SSD，显示出有希望的结果，相对于现有技术的两阶段方法，产生更快的探测器，精度在10-40％之内。
Non-Stationary Texture Synthesis by Adversarial Expansion
1.
In our approach, a dedicated GAN must be trained for each input exemplar, which takes considerable computational resources.
在我们的方法中，必须针对每个输入示例训练专用GAN，这需要相当多的计算资源。
