Shortcuts

Tutorial 6: Customize Losses

MMDetection provides users with different loss functions. But the default configuration may be not applicable for different datasets or models, so users may want to modify a specific loss to adapt the new situation.

This tutorial first elaborate the computation pipeline of losses, then give some instructions about how to modify each step. The modification can be categorized as tweaking and weighting.

Computation pipeline of a loss

Given the input prediction and target, as well as the weights, a loss function maps the input tensor to the final loss scalar. The mapping can be divided into four steps:

  1. Get element-wise or sample-wise loss by the loss kernel function.

  2. Weighting the loss with a weight tensor element-wisely.

  3. Reduce the loss tensor to a scalar.

  4. Weighting the loss with a scalar.

Tweaking loss

Tweaking a loss is more related with step 1, 3, 4, and most modifications can be specified in the config. Here we take Focal Loss (FL) as an example. The following code sniper are the construction method and config of FL respectively, they are actually one to one correspondence.

@LOSSES.register_module()
class FocalLoss(nn.Module):

    def __init__(self,
                 use_sigmoid=True,
                 gamma=2.0,
                 alpha=0.25,
                 reduction='mean',
                 loss_weight=1.0):
loss_cls=dict(
    type='FocalLoss',
    use_sigmoid=True,
    gamma=2.0,
    alpha=0.25,
    loss_weight=1.0)

Tweaking hyper-parameters (step 1)

gamma and beta are two hyper-parameters in the Focal Loss. Say if we want to change the value of gamma to be 1.5 and alpha to be 0.5, then we can specify them in the config as follows:

loss_cls=dict(
    type='FocalLoss',
    use_sigmoid=True,
    gamma=1.5,
    alpha=0.5,
    loss_weight=1.0)

Tweaking the way of reduction (step 3)

The default way of reduction is mean for FL. Say if we want to change the reduction from mean to sum, we can specify it in the config as follows:

loss_cls=dict(
    type='FocalLoss',
    use_sigmoid=True,
    gamma=2.0,
    alpha=0.25,
    loss_weight=1.0,
    reduction='sum')

Tweaking loss weight (step 4)

The loss weight here is a scalar which controls the weight of different losses in multi-task learning, e.g. classification loss and regression loss. Say if we want to change to loss weight of classification loss to be 0.5, we can specify it in the config as follows:

loss_cls=dict(
    type='FocalLoss',
    use_sigmoid=True,
    gamma=2.0,
    alpha=0.25,
    loss_weight=0.5)

Weighting loss (step 2)

Weighting loss means we re-weight the loss element-wisely. To be more specific, we multiply the loss tensor with a weight tensor which has the same shape. As a result, different entries of the loss can be scaled differently, and so called element-wisely. The loss weight varies across different models and highly context related, but overall there are two kinds of loss weights, label_weights for classification loss and bbox_weights for bbox regression loss. You can find them in the get_target method of the corresponding head. Here we take ATSSHead as an example, which inherit AnchorHead but overwrite its get_targets method which yields different label_weights and bbox_weights.

class ATSSHead(AnchorHead):

    ...

    def get_targets(self,
                    anchor_list,
                    valid_flag_list,
                    gt_bboxes_list,
                    img_metas,
                    gt_bboxes_ignore_list=None,
                    gt_labels_list=None,
                    label_channels=1,
                    unmap_outputs=True):
Read the Docs v: latest
Versions
latest
stable
v2.17.0
v2.16.0
v2.15.1
v2.15.0
v2.14.0
v2.13.0
v2.12.0
v2.11.0
v2.10.0
v2.9.0
v2.8.0
v2.7.0
v2.6.0
v2.5.0
v2.4.0
v2.3.0
v2.2.1
v2.2.0
v2.1.0
v2.0.0
v1.2.0
Downloads
pdf
html
epub
On Read the Docs
Project Home
Builds

Free document hosting provided by Read the Docs.