Tutorial 1: Finetuning Models

Detectors pre-trained on the COCO dataset can serve as a good pre-trained model for other datasets, e.g., CityScapes and KITTI Dataset. This tutorial provides instruction for users to use the models provided in the Model Zoo for other datasets to obtain better performance.

There are two steps to finetune a model on a new dataset.

Take the finetuning process on Cityscapes Dataset as an example, the users need to modify five parts in the config.

Inherit base configs

To release the burden and reduce bugs in writing the whole configs, MMDetection V2.0 support inheriting configs from multiple existing configs. To finetune a Mask RCNN model, the new config needs to inherit _base_/models/mask_rcnn_r50_fpn.py to build the basic structure of the model. To use the Cityscapes Dataset, the new config can also simply inherit _base_/datasets/cityscapes_instance.py. For runtime settings such as training schedules, the new config needs to inherit _base_/default_runtime.py. This configs are in the configs directory and the users can also choose to write the whole contents rather than use inheritance.

_base_ = [
    '../_base_/models/mask_rcnn_r50_fpn.py',
    '../_base_/datasets/cityscapes_instance.py', '../_base_/default_runtime.py'
]

Modify head

Then the new config needs to modify the head according to the class numbers of the new datasets. By only changing num_classes in the roi_head, the weights of the pre-trained models are mostly reused except the final prediction head.

model = dict(
    pretrained=None,
    roi_head=dict(
        bbox_head=dict(
            type='Shared2FCBBoxHead',
            in_channels=256,
            fc_out_channels=1024,
            roi_feat_size=7,
            num_classes=8,
            bbox_coder=dict(
                type='DeltaXYWHBBoxCoder',
                target_means=[0., 0., 0., 0.],
                target_stds=[0.1, 0.1, 0.2, 0.2]),
            reg_class_agnostic=False,
            loss_cls=dict(
                type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
            loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),
        mask_head=dict(
            type='FCNMaskHead',
            num_convs=4,
            in_channels=256,
            conv_out_channels=256,
            num_classes=8,
            loss_mask=dict(
                type='CrossEntropyLoss', use_mask=True, loss_weight=1.0))))

Modify dataset

The users may also need to prepare the dataset and write the configs about dataset. MMDetection V2.0 already support VOC, WIDER FACE, COCO and Cityscapes Dataset.

Modify training schedule

The finetuning hyperparameters vary from the default schedule. It usually requires smaller learning rate and less training epochs

# optimizer
# lr is set for a batch size of 8
optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
    policy='step',
    warmup='linear',
    warmup_iters=500,
    warmup_ratio=0.001,
    # [7] yields higher performance than [6]
    step=[7])
total_epochs = 8  # actual epoch = 8 * 8 = 64
log_config = dict(interval=100)

Use pre-trained model

To use the pre-trained model, the new config add the link of pre-trained models in the load_from. The users might need to download the model weights before training to avoid the download time during training.

load_from = 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/mmdetection/models/mask_rcnn_r50_fpn_2x_20181010-41d35c05.pth'  # noqa