how to use augmentation in train_net.py ?

[leesangjoon1]

I wonder how to use augmentation in train_net.py

and if the augmentation is applied, where can I check the which augmentation is applied?

and can I use cutmix augmentation in detectron2?

[AKMourato]

Augmentations are already applied in the provided training scripts.
You can check at the runtime for instance:

[08/16 20:26:52 d2.data.datasets.coco]: Loaded 1408 images in COCO format from ./2011/train/instances_images.json
[08/16 20:26:52 d2.data.build]: Removed 0 images with no usable annotations. 1408 images left.
[08/16 20:26:52 d2.data.build]: Distribution of instances among all 1 categories:
|  category  | #instances   |
|:----------:|:-------------|
|    ssg     | 9363         |
|            |              |
[08/16 20:26:52 d2.data.dataset_mapper]: [DatasetMapper] Augmentations used in training: [ResizeShortestEdge(short_edge_length=(2048, 2048), max_size=4096, sample_style='choice'), RandomFlip()]
[08/16 20:26:52 d2.data.build]: Using training sampler TrainingSampler
[08/16 20:26:52 d2.data.common]: Serializing 1408 elements to byte tensors and concatenating them all ...
[08/16 20:26:52 d2.data.common]: Serialized dataset takes 12.88 MiB

Above is an example of the augmentations of the DefaultTrainer - resize and random flip.
Check https://detectron2.readthedocs.io/en/latest/tutorials/augmentation.html to write custom ones.

[leesangjoon1]

but in datasetmapper, there are so many kinds of augmentations. so I want to know how to use them all in train_net.py