htcr

Hi, first of all, I think you may want to try to scale the images such that they are 1.0m / pixel. That's how these checkpoints are trained. Can you...

I think the CityScale setup defaults to 512x512 patch size. Can you try that? Batch size depends on your GPU memory, I think you can start with the largest batch...

I think our released model takes 1.0m/pixel images. Can you try resizing your images to that resolution? Also, have you fine-tuned on your own dataset? How large was your dataset?

Also did you correctly load the pre-trained SAM ckpts?

Hi, I think if you are fine-tuning from the original SAM ckpt (not the ones I released), resolution is less crucial. How does the images look like in general? The...

Hi, does your dataset contain ground-truth graphs? You can process that graph into the same format as used in this codebase, then it shall generate keypoint masks for you.

For example, you can refer to the generate_labels.py under cityscales dir. From line 82, it reads the GT graph - and it reveals the gt graph format, basically an adjacency...

Ok, if you were trying to re-train/fine-tune on your dataset, you do need ground-truth graph. Otherwise if you just want to run our checkpoint on some images, you can just...

> > Hello, I try to start inference on my own images via command: `python3 inferencer.py --config=config/toponet_vitb_512_ekb.yaml --checkpoint=lightning_logs/vhfsw197/checkpoints/epoch=9-step=25000.ckpt` > > _and e.t.c._ > > Please, can you explane me what...

> 方便提供一个requirements安装环境，不然会有各种报错， 例如： File "/home/sam_road-main/model.py", line 102, in forward src_features = point_features[batch_indices, pairs[:, :, 0]] IndexError: tensors used as indices must be long, byte or bool tensors > > python3.8/site-packages/torch/nn/modules/transformer.py",...