(ckpt_path, model_name, N=512, model_type='bacon', hidden_layers=8,
hidden_size=256, output_layers=[1, 2, 4, 8],
return_sdf=False, adaptive=True)
| 14 | |
| 15 | |
| 16 | def export_model(ckpt_path, model_name, N=512, model_type='bacon', hidden_layers=8, |
| 17 | hidden_size=256, output_layers=[1, 2, 4, 8], |
| 18 | return_sdf=False, adaptive=True): |
| 19 | |
| 20 | # the network has 4 output levels of detail |
| 21 | num_outputs = len(output_layers) |
| 22 | max_frequency = 3*(32,) |
| 23 | |
| 24 | # load model |
| 25 | with utils.HiddenPrint(): |
| 26 | model = modules.MultiscaleBACON(3, hidden_size, 1, |
| 27 | hidden_layers=hidden_layers, |
| 28 | bias=True, |
| 29 | frequency=max_frequency, |
| 30 | quantization_interval=np.pi, |
| 31 | is_sdf=True, |
| 32 | output_layers=output_layers, |
| 33 | reuse_filters=True) |
| 34 | |
| 35 | ckpt = torch.load(ckpt_path) |
| 36 | model.load_state_dict(ckpt) |
| 37 | model.cuda() |
| 38 | |
| 39 | if not adaptive: |
| 40 | # extracts separate meshes for each scale |
| 41 | generate_mesh(model, N, return_sdf, num_outputs, model_name) |
| 42 | |
| 43 | else: |
| 44 | # extracts single-scale output |
| 45 | generate_mesh_adaptive(model, model_name) |
| 46 | |
| 47 | |
| 48 | def generate_mesh(model, N, return_sdf=False, num_outputs=4, model_name='model'): |
no test coverage detected