| 134 | |
| 135 | |
| 136 | def render_path(render_poses, hwf, chunk, render_kwargs, gt_imgs=None, savedir=None, render_factor=0): |
| 137 | |
| 138 | H, W, focal = hwf |
| 139 | |
| 140 | if render_factor!=0: |
| 141 | # Render downsampled for speed |
| 142 | H = H//render_factor |
| 143 | W = W//render_factor |
| 144 | focal = focal/render_factor |
| 145 | |
| 146 | rgbs = [] |
| 147 | disps = [] |
| 148 | uncerts = [] |
| 149 | |
| 150 | t = time.time() |
| 151 | for i, c2w in enumerate(tqdm(render_poses)): |
| 152 | print(i, time.time() - t) |
| 153 | t = time.time() |
| 154 | rgb, disp, acc, uncert, alpha, _ = render(H, W, focal, chunk=chunk, c2w=c2w[:3,:4], **render_kwargs) |
| 155 | rgbs.append(rgb.cpu().numpy()) |
| 156 | disps.append(disp.cpu().numpy()) |
| 157 | uncerts.append(uncert.cpu().numpy()) |
| 158 | |
| 159 | """ |
| 160 | if gt_imgs is not None and render_factor==0: |
| 161 | p = -10. * np.log10(np.mean(np.square(rgb.cpu().numpy() - gt_imgs[i]))) |
| 162 | print(p) |
| 163 | """ |
| 164 | |
| 165 | if savedir is not None: |
| 166 | rgb8 = to8b(rgbs[-1]) |
| 167 | filename = os.path.join(savedir, '{:03d}.png'.format(i)) |
| 168 | imageio.imwrite(filename, rgb8) |
| 169 | |
| 170 | uncert8 = to8b(uncerts[-1]) |
| 171 | filename = os.path.join(savedir, '{:03d}_uncert.png'.format(i)) |
| 172 | imageio.imwrite(filename, uncert8) |
| 173 | |
| 174 | torch.cuda.empty_cache() |
| 175 | |
| 176 | rgbs = np.stack(rgbs, 0) |
| 177 | disps = np.stack(disps, 0) |
| 178 | uncerts = np.stack(uncerts, 0) |
| 179 | |
| 180 | return rgbs, disps, uncerts, None |
| 181 | |
| 182 | |
| 183 | def create_nerf(args): |