Loss converged upon unfreezing aggregator but results bad
Hi, I recently finetuned VGGT's camera head on COLMAP data ; While unfreezing the camera head resulted in convergence, upon inferencing, the results are worse than the pretrained VGGT's output.
What are some things I should keep an eye on?
Hi, I’m facing the same issue. Have you found a solution?
Hi can you try to use a smaller lr, e.g., 10% of the original one?
Hello, can you share how to fine-tune VGGT using COLMAP data? :)
Hi, I recently finetuned VGGT's camera head on COLMAP data ; While unfreezing the camera head resulted in convergence, upon inferencing, the results are worse than the pretrained VGGT's output.嗨,我最近在 COLMAP 数据上微调了 VGGT 的相机头;在解冻相机头后虽然实现了收敛,但在推理时,结果比预训练的 VGGT 输出更差。
What are some things I should keep an eye on?我应该关注哪些事项?