Dmytro Mishkin

Actually, this is enough for the package, but not data and config files, let me update them as well

Yes. It actually requires much more work than I thought, because many weights loading and import operations relies on the editable install - and not only limap, but submodules as...

@B1ueber2y while you are at here, is there an easy way of exporting the optimized BA results instead of `limapio.save_npy(os.path.join(global_dir_save, "imagecols_optimized.npy"), new_imagecols)` to COLMAP?

You too! Thank you!

Even with this fix, we have an error: ``` y = torch.multinomial(x, min(max_num_cmp,x.shape[-2]), replacement=False) RuntimeError: cannot sample n_sample > prob_dist.size(-1) samples without replacement ```

@qubvel I believe I shared this in: `Parameters which did not receive grad for rank 3: neck.fusion_stage.layers.0.residual_layer1.convolution2.bias, neck.fusion_stage.layers.0.residual_layer1.convolution2.weight, neck.fusion_stage.layers.0.residual_layer1.convolution1.bias, neck.fusion_stage.layers.0.residual_layer1.convolution1.weight` That is a quote from the error crash message I...

@qubvel good point about the backbone. Probably because I have trained with a frozen backbone, which is kind of common. And about the backbone removing unused params there would probably...

+1

@lhoestq do you mean to add the decoder, and exr extension to the image format? Yes, that probably would be better

@AntoineSimoulin I do understand it. However, have you considered that both links that torchvision used before, are the same thing that I just created- personal websites under the university domain....