Could you release AutoPET finetuning weight?

[bbuok-neu]

Thank you for your work.

I've been working on PET tumor detection research recently, and my dataset is quite small. The weights you fine-tuned on AutoPET are very important to me, and I'd like to fine-tune on top of them.

I also have a question: my GPU only has 11GB of VRAM. Should I use the S or B model?

[Ziyan-Huang]

Thanks for the reminder! We’ll try to release some weights, including AutoPET, within 1–2 weeks and will let you know once they’re available. In the meantime, we suggest fine-tuning from our released weights on TotalSeg, as in our experience, pretraining on a larger and more general dataset often transfers better than weights from a smaller, task-specific dataset like AutoPET. I’d also recommend trying at least the B model, as the S model may not benefit much from pretrained weights.

[bbuok-neu]

Thanks for the reminder! We’ll try to release some weights, including AutoPET, within 1–2 weeks and will let you know once they’re available. In the meantime, we suggest fine-tuning from our released weights on TotalSeg, as in our experience, pretraining on a larger and more general dataset often transfers better than weights from a smaller, task-specific dataset like AutoPET. I’d also recommend trying at least the B model, as the S model may not benefit much from pretrained weights.感谢提醒！我们将在 1-2 周内尝试发布一些权重，包括 AutoPET，一旦可用我们会通知您。在此期间，我们建议从我们发布的 TotalSeg 权重中进行微调，因为根据我们的经验，在更大、更通用的数据集上进行预训练通常比来自较小、特定任务数据集（如 AutoPET）的权重转移效果更好。我还建议至少尝试 B 模型，因为 S 模型可能不会从预训练权重中获得太多好处。

Thank you for the explanation. I look forward to the release.
谢谢回复，期待一下。

[Ziyan-Huang]

Hi! I’ve uploaded my previously fine-tuned checkpoints for AutoPET here (Google Drive → AutoPET folder): https://drive.google.com/drive/folders/1VYfpPLANIrlQdR3HnNjZMtJBx7OPGi64
Hope this helps!

[bbuok-neu]

Thanks so much! Wishing you all the best with your work.