Debug

[KyleRoss-rice]

Fix for "off-by-one" issue in the main loop

for i in range(opts.iterations):
    //...

    if i % 1 == 0:
        print(f'{os.path.basename(opts.input_pc)}; iter: {i} out of: {opts.iterations}; loss: {loss.item():.4f};'
                f' sample count: {num_samples}; time: {end_time - start_time:.2f}')
    if i % opts.export_interval == 0 and i > 0:
        print('exporting reconstruction... current LR: {}'.format(optimizer.param_groups[0]['lr']))
        with torch.no_grad():
            part_mesh.export(os.path.join(opts.save_path, f'recon_iter_{i}.obj'))

    //...

if opts.iterations == 100, for example, then the range for i is [0, 99].
On the final iteration, iter: 99 out of: 100 is printed.

if opts.export_interval == 100, then i % opts.export_interval == 0 is always false, since i never reaches a multiple of 100. No reconstructed mesh is exported.