Which data set are the pretrained weight files trained on?

[GZhengyang]

Are the pre-training weights provided directly trained based on the 258 image-label pairs of the trainset provided by the competition party, or are they trained based on the coco data set or PASCAL-5i data set in DLAM?

[cliffbb]

@GZhengyang Yes, the baseline is pre-trained only on the 258 image-label pairs of the trainset. But, in addition to the 256 images, you can train your backbone on any dataset except remote sensing datasets.