17CVPR| (Oral) Focal Loss for Dense Object Detection

[XFeiF]

[Paper]
[Pytorch Code]

Main idea:

To solve the extreme foreground-background class imbalance problem in one-stage object detection frameworks, the authors down-weights the loss assigned to well-classified examples by proposing Focal Loss, which adds a factor to the standard cross entropy criterion.

How focal loss works ?

As shown in the figure above, the factor will lower the standard cross entropy loss. Suppose gamma is 2, an example classified with p_t = 0.9 would have 100× lower loss compared with CE and with p_t ≈ 0.968 it would have 1000× lower loss. This in turn increases the importance of correcting misclassified examples (whose loss is scaled down by at most 4× for p_t ≤ .5 and gamma = 2).

[XFeiF]

Several keynotes for me

The results of class imbalance

• training is inefficient as most locations are easy negatives that contribute no useful learning signal;
• en masse, the easy negatives can overwhelm training and lead to degenerate models.

Two properties of the focal loss

• When an example is misclassified and p_t is small, the modulating factor is near 1 and the loss is unaffected. As p_t -> 1, the factor goes to 0 and the loss for well-classified examples is down-weighted.
• The focusing parameter gamma smoothly adjusts the rate at which easy examples are down-weighted.

How two-stage detectors address class imbalance?

• A two-stage cascade. The first cascade stage is an object proposal mechanism that reduces the nearly infinite set of possible object locations down to one or two thousand.
• Biased minibatch sampling. When training the second stage, biased sampling is typically used to construct minibatches that contain, for instance, a 1:3 ratio of positive to negative examples.