File size: 24,363 Bytes
71fa249 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 |
{
"cells": [
{
"cell_type": "markdown",
"metadata": {
"id": "EZ_wkNVdTz-C"
},
"source": [
"# Let's train HyperNeRF!\n",
"\n",
"**Author**: [Keunhong Park](https://keunhong.com)\n",
"\n",
"[[Project Page](https://hypernerf.github.io)]\n",
"[[Paper](https://arxiv.org/abs/2106.13228)]\n",
"[[GitHub](https://github.com/google/hypernerf)]\n",
"\n",
"This notebook provides an demo for training HyperNeRF.\n",
"\n",
"### Instructions\n",
"\n",
"1. Convert a video into our dataset format using the Nerfies [dataset processing notebook](https://colab.sandbox.google.com/github/google/nerfies/blob/main/notebooks/Nerfies_Capture_Processing.ipynb).\n",
"2. Set the `data_dir` below to where you saved the dataset.\n",
"3. Come back to this notebook to train HyperNeRF.\n",
"\n",
"\n",
"### Notes\n",
" * To accomodate the limited compute power of Colab runtimes, this notebook defaults to a \"toy\" version of our method. The number of samples have been reduced and the elastic regularization turned off.\n",
"\n",
" * To train a high-quality model, please look at the CLI options we provide in the [Github repository](https://github.com/google/hypernerf).\n",
"\n",
"\n",
"\n",
" * Please report issues on the [GitHub issue tracker](https://github.com/google/hypernerf/issues).\n",
"\n",
"\n",
"If you find this work useful, please consider citing:\n",
"```bibtex\n",
"@article{park2021hypernerf\n",
" author = {Park, Keunhong and Sinha, Utkarsh and Hedman, Peter and Barron, Jonathan T. and Bouaziz, Sofien and Goldman, Dan B and Martin-Brualla, Ricardo and Seitz, Steven M.},\n",
" title = {HyperNeRF: A Higher-Dimensional Representation for Topologically Varying Neural Radiance Fields},\n",
" journal = {arXiv preprint arXiv:2106.13228},\n",
" year = {2021},\n",
"}\n",
"```\n"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "OlW1gF_djH6H"
},
"source": [
"## Environment Setup"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "lMGu9ctBT-MD",
"outputId": "41a8dd06-943a-4820-c2cf-e98a25a167e7"
},
"outputs": [],
"source": [
"#!wget https://raw.githubusercontent.com/google/hypernerf/main/requirements.txt\n",
"!wget https://raw.githubusercontent.com/xieyizheng/hypernerf/main/requirements.txt\n",
"!python --version\n",
"!pip install -r requirements.txt"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "ns2J1yBAsYgt",
"outputId": "6c73222d-8643-4fe7-8f90-1b2ab79465df"
},
"outputs": [],
"source": [
"!nvidia-smi"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"\n",
"#if only freshly installed the requirements, recommend to restart the runtime!\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "zGJux-m5Xp3Z",
"outputId": "58e386b0-44be-4741-8dbf-43cb46dade40"
},
"outputs": [],
"source": [
"# @title Configure notebook runtime\n",
"#***only gpu works\n",
"# @markdown If you would like to use a GPU runtime instead, change the runtime type by going to `Runtime > Change runtime type`. \n",
"# @markdown You will have to use a smaller batch size on GPU.\n",
"import jax\n",
"#jax.config.update('jax_platform_name', 'gpu')\n",
"runtime_type = 'gpu' # @param ['gpu', 'tpu']\n",
"if runtime_type == 'tpu':\n",
" import jax.tools.colab_tpu\n",
" jax.tools.colab_tpu.setup_tpu()\n",
"\n",
"print('Detected Devices:', jax.devices())"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"cellView": "form",
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "afUtLfRWULEi",
"outputId": "2919d242-fa49-447d-934e-877fbb42e5de"
},
"outputs": [],
"source": [
"# @title Mount Google Drive\n",
"# @markdown Mount Google Drive onto `/content/gdrive`. You can skip this if running locally.\n",
"\n",
"#use accordingly, if local, comment this out\n",
"from google.colab import drive\n",
"drive.mount('/content/gdrive')"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "ENOfbG3AkcVN"
},
"outputs": [],
"source": [
"# @title Define imports and utility functions.\n",
"\n",
"import jax\n",
"from jax.config import config as jax_config\n",
"import jax.numpy as jnp\n",
"from jax import grad, jit, vmap\n",
"from jax import random\n",
"\n",
"import flax\n",
"import flax.linen as nn\n",
"from flax import jax_utils\n",
"from flax import optim\n",
"from flax.metrics import tensorboard\n",
"from flax.training import checkpoints\n",
"#jax_config.enable_omnistaging() # Linen requires enabling omnistaging\n",
"\n",
"from absl import logging\n",
"from io import BytesIO\n",
"import random as pyrandom\n",
"import numpy as np\n",
"import PIL\n",
"import IPython\n",
"\n",
"\n",
"# Monkey patch logging.\n",
"def myprint(msg, *args, **kwargs):\n",
" print(msg % args)\n",
"\n",
"logging.info = myprint \n",
"logging.warn = myprint\n",
"logging.error = myprint\n",
"\n",
"\n",
"def show_image(image, fmt='png'):\n",
" image = image_utils.image_to_uint8(image)\n",
" f = BytesIO()\n",
" PIL.Image.fromarray(image).save(f, fmt)\n",
" IPython.display.display(IPython.display.Image(data=f.getvalue()))\n",
"\n"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "wW7FsSB-jORB"
},
"source": [
"## Configuration"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"cellView": "form",
"colab": {
"base_uri": "https://localhost:8080/",
"height": 1000
},
"id": "rz7wRm7YT9Ka",
"outputId": "8d185175-603a-4903-ad18-7626eb8d1d91"
},
"outputs": [],
"source": [
"# @title Model and dataset configuration\n",
"\n",
"from pathlib import Path\n",
"from pprint import pprint\n",
"import gin\n",
"from IPython.display import display, Markdown\n",
"\n",
"from hypernerf import models\n",
"from hypernerf import modules\n",
"from hypernerf import warping\n",
"from hypernerf import datasets\n",
"from hypernerf import configs\n",
"\n",
"\n",
"# @markdown The working directory.\n",
"train_dir = '/content/gdrive/My Drive/nerfies/hypernerf_experiments/hand/exp1' # @param {type: \"string\"}\n",
"# @markdown The directory to the dataset capture.\n",
"data_dir = '/content/gdrive/My Drive/nerfies/captures/hand' # @param {type: \"string\"}\n",
"\n",
"# @markdown Training configuration.\n",
"max_steps = 200000 # @param {type: 'number'}\n",
"batch_size = 2048 # @param {type: 'number'}\n",
"image_scale = 8 # @param {type: 'number'}\n",
"\n",
"# @markdown Model configuration.\n",
"use_viewdirs = True #@param {type: 'boolean'}\n",
"use_appearance_metadata = True #@param {type: 'boolean'}\n",
"num_coarse_samples = 64 # @param {type: 'number'}\n",
"num_fine_samples = 64 # @param {type: 'number'}\n",
"\n",
"# @markdown Deformation configuration.\n",
"use_warp = True #@param {type: 'boolean'}\n",
"warp_field_type = '@SE3Field' #@param['@SE3Field', '@TranslationField']\n",
"warp_min_deg = 0 #@param{type:'number'}\n",
"warp_max_deg = 6 #@param{type:'number'}\n",
"\n",
"# @markdown Hyper-space configuration.\n",
"hyper_num_dims = 8 #@param{type:'number'}\n",
"hyper_point_min_deg = 0 #@param{type:'number'}\n",
"hyper_point_max_deg = 1 #@param{type:'number'}\n",
"hyper_slice_method = 'bendy_sheet' #@param['none', 'axis_aligned_plane', 'bendy_sheet']\n",
"\n",
"\n",
"checkpoint_dir = Path(train_dir, 'checkpoints')\n",
"checkpoint_dir.mkdir(exist_ok=True, parents=True)\n",
"\n",
"config_str = f\"\"\"\n",
"DELAYED_HYPER_ALPHA_SCHED = {{\n",
" 'type': 'piecewise',\n",
" 'schedules': [\n",
" (1000, ('constant', 0.0)),\n",
" (0, ('linear', 0.0, %hyper_point_max_deg, 10000))\n",
" ],\n",
"}}\n",
"\n",
"ExperimentConfig.image_scale = {image_scale}\n",
"ExperimentConfig.datasource_cls = @NerfiesDataSource\n",
"NerfiesDataSource.data_dir = '{data_dir}'\n",
"NerfiesDataSource.image_scale = {image_scale}\n",
"\n",
"NerfModel.use_viewdirs = {int(use_viewdirs)}\n",
"NerfModel.use_rgb_condition = {int(use_appearance_metadata)}\n",
"NerfModel.num_coarse_samples = {num_coarse_samples}\n",
"NerfModel.num_fine_samples = {num_fine_samples}\n",
"\n",
"NerfModel.use_viewdirs = True\n",
"NerfModel.use_stratified_sampling = True\n",
"NerfModel.use_posenc_identity = False\n",
"NerfModel.nerf_trunk_width = 128\n",
"NerfModel.nerf_trunk_depth = 8\n",
"\n",
"TrainConfig.max_steps = {max_steps}\n",
"TrainConfig.batch_size = {batch_size}\n",
"TrainConfig.print_every = 100\n",
"TrainConfig.use_elastic_loss = False\n",
"TrainConfig.use_background_loss = False\n",
"\n",
"# Warp configs.\n",
"warp_min_deg = {warp_min_deg}\n",
"warp_max_deg = {warp_max_deg}\n",
"NerfModel.use_warp = {use_warp}\n",
"SE3Field.min_deg = %warp_min_deg\n",
"SE3Field.max_deg = %warp_max_deg\n",
"SE3Field.use_posenc_identity = False\n",
"NerfModel.warp_field_cls = @SE3Field\n",
"\n",
"TrainConfig.warp_alpha_schedule = {{\n",
" 'type': 'linear',\n",
" 'initial_value': {warp_min_deg},\n",
" 'final_value': {warp_max_deg},\n",
" 'num_steps': {int(max_steps*0.8)},\n",
"}}\n",
"\n",
"# Hyper configs.\n",
"hyper_num_dims = {hyper_num_dims}\n",
"hyper_point_min_deg = {hyper_point_min_deg}\n",
"hyper_point_max_deg = {hyper_point_max_deg}\n",
"\n",
"NerfModel.hyper_embed_cls = @hyper/GLOEmbed\n",
"hyper/GLOEmbed.num_dims = %hyper_num_dims\n",
"NerfModel.hyper_point_min_deg = %hyper_point_min_deg\n",
"NerfModel.hyper_point_max_deg = %hyper_point_max_deg\n",
"\n",
"TrainConfig.hyper_alpha_schedule = %DELAYED_HYPER_ALPHA_SCHED\n",
"\n",
"hyper_sheet_min_deg = 0\n",
"hyper_sheet_max_deg = 6\n",
"HyperSheetMLP.min_deg = %hyper_sheet_min_deg\n",
"HyperSheetMLP.max_deg = %hyper_sheet_max_deg\n",
"HyperSheetMLP.output_channels = %hyper_num_dims\n",
"\n",
"NerfModel.hyper_slice_method = '{hyper_slice_method}'\n",
"NerfModel.hyper_sheet_mlp_cls = @HyperSheetMLP\n",
"NerfModel.hyper_use_warp_embed = True\n",
"\n",
"TrainConfig.hyper_sheet_alpha_schedule = ('constant', %hyper_sheet_max_deg)\n",
"\"\"\"\n",
"\n",
"gin.parse_config(config_str)\n",
"\n",
"config_path = Path(train_dir, 'config.gin')\n",
"with open(config_path, 'w') as f:\n",
" logging.info('Saving config to %s', config_path)\n",
" f.write(config_str)\n",
"\n",
"exp_config = configs.ExperimentConfig()\n",
"train_config = configs.TrainConfig()\n",
"eval_config = configs.EvalConfig()\n",
"\n",
"display(Markdown(\n",
" gin.config.markdown(gin.config_str())))"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"cellView": "form",
"colab": {
"base_uri": "https://localhost:8080/",
"height": 533
},
"id": "r872r6hiVUVS",
"outputId": "f8794983-1165-4e93-8236-6cac48bbd552"
},
"outputs": [],
"source": [
"# @title Create datasource and show an example.\n",
"\n",
"from hypernerf import datasets\n",
"from hypernerf import image_utils\n",
"\n",
"dummy_model = models.NerfModel({}, 0, 0)\n",
"datasource = exp_config.datasource_cls(\n",
" image_scale=exp_config.image_scale,\n",
" random_seed=exp_config.random_seed,\n",
" # Enable metadata based on model needs.\n",
" use_warp_id=dummy_model.use_warp,\n",
" use_appearance_id=(\n",
" dummy_model.nerf_embed_key == 'appearance'\n",
" or dummy_model.hyper_embed_key == 'appearance'),\n",
" use_camera_id=dummy_model.nerf_embed_key == 'camera',\n",
" use_time=dummy_model.warp_embed_key == 'time')\n",
"\n",
"show_image(datasource.load_rgb(datasource.train_ids[0]))"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "XC3PIY74XB05",
"outputId": "b2f57210-07ff-49c5-b51b-87864d4a1f17"
},
"outputs": [],
"source": [
"# @title Create training iterators\n",
"\n",
"devices = jax.local_devices()\n",
"\n",
"train_iter = datasource.create_iterator(\n",
" datasource.train_ids,\n",
" flatten=True,\n",
" shuffle=True,\n",
" batch_size=train_config.batch_size,\n",
" prefetch_size=3,\n",
" shuffle_buffer_size=train_config.shuffle_buffer_size,\n",
" devices=devices,\n",
")\n",
"\n",
"def shuffled(l):\n",
" import random as r\n",
" import copy\n",
" l = copy.copy(l)\n",
" r.shuffle(l)\n",
" return l\n",
"\n",
"train_eval_iter = datasource.create_iterator(\n",
" shuffled(datasource.train_ids), batch_size=0, devices=devices)\n",
"val_eval_iter = datasource.create_iterator(\n",
" shuffled(datasource.val_ids), batch_size=0, devices=devices)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "erY9l66KjYYW"
},
"source": [
"## Training"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "nZnS8BhcXe5E",
"outputId": "980fda5d-863e-4b7e-d3dd-aff48dd4950a"
},
"outputs": [],
"source": [
"# @title Initialize model\n",
"# @markdown Defines the model and initializes its parameters.\n",
"\n",
"from flax.training import checkpoints\n",
"from hypernerf import models\n",
"from hypernerf import model_utils\n",
"from hypernerf import schedules\n",
"from hypernerf import training\n",
"\n",
"# @markdown Restore a checkpoint if one exists.\n",
"restore_checkpoint = True # @param{type:'boolean'}\n",
"\n",
"\n",
"rng = random.PRNGKey(exp_config.random_seed)\n",
"np.random.seed(exp_config.random_seed + jax.process_index())\n",
"devices_to_use = jax.devices()\n",
"\n",
"learning_rate_sched = schedules.from_config(train_config.lr_schedule)\n",
"nerf_alpha_sched = schedules.from_config(train_config.nerf_alpha_schedule)\n",
"warp_alpha_sched = schedules.from_config(train_config.warp_alpha_schedule)\n",
"elastic_loss_weight_sched = schedules.from_config(\n",
"train_config.elastic_loss_weight_schedule)\n",
"hyper_alpha_sched = schedules.from_config(train_config.hyper_alpha_schedule)\n",
"hyper_sheet_alpha_sched = schedules.from_config(\n",
" train_config.hyper_sheet_alpha_schedule)\n",
"\n",
"rng, key = random.split(rng)\n",
"params = {}\n",
"model, params['model'] = models.construct_nerf(\n",
" key,\n",
" batch_size=train_config.batch_size,\n",
" embeddings_dict=datasource.embeddings_dict,\n",
" near=datasource.near,\n",
" far=datasource.far)\n",
"\n",
"optimizer_def = optim.Adam(learning_rate_sched(0))\n",
"optimizer = optimizer_def.create(params)\n",
"\n",
"state = model_utils.TrainState(\n",
" optimizer=optimizer,\n",
" nerf_alpha=nerf_alpha_sched(0),\n",
" warp_alpha=warp_alpha_sched(0),\n",
" hyper_alpha=hyper_alpha_sched(0),\n",
" hyper_sheet_alpha=hyper_sheet_alpha_sched(0))\n",
"scalar_params = training.ScalarParams(\n",
" learning_rate=learning_rate_sched(0),\n",
" elastic_loss_weight=elastic_loss_weight_sched(0),\n",
" warp_reg_loss_weight=train_config.warp_reg_loss_weight,\n",
" warp_reg_loss_alpha=train_config.warp_reg_loss_alpha,\n",
" warp_reg_loss_scale=train_config.warp_reg_loss_scale,\n",
" background_loss_weight=train_config.background_loss_weight,\n",
" hyper_reg_loss_weight=train_config.hyper_reg_loss_weight)\n",
"\n",
"if restore_checkpoint:\n",
" logging.info('Restoring checkpoint from %s', checkpoint_dir)\n",
" state = checkpoints.restore_checkpoint(checkpoint_dir, state)\n",
"step = state.optimizer.state.step + 1\n",
"state = jax_utils.replicate(state, devices=devices)\n",
"del params"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "at2CL5DRZ7By"
},
"outputs": [],
"source": [
"# @title Define pmapped functions\n",
"# @markdown This parallelizes the training and evaluation step functions using `jax.pmap`.\n",
"\n",
"import functools\n",
"from hypernerf import evaluation\n",
"\n",
"\n",
"def _model_fn(key_0, key_1, params, rays_dict, extra_params):\n",
" out = model.apply({'params': params},\n",
" rays_dict,\n",
" extra_params=extra_params,\n",
" rngs={\n",
" 'coarse': key_0,\n",
" 'fine': key_1\n",
" },\n",
" mutable=False)\n",
" return jax.lax.all_gather(out, axis_name='batch')\n",
"\n",
"pmodel_fn = jax.pmap(\n",
" # Note rng_keys are useless in eval mode since there's no randomness.\n",
" _model_fn,\n",
" in_axes=(0, 0, 0, 0, 0), # Only distribute the data input.\n",
" devices=devices_to_use,\n",
" axis_name='batch',\n",
")\n",
"\n",
"render_fn = functools.partial(evaluation.render_image,\n",
" model_fn=pmodel_fn,\n",
" device_count=len(devices),\n",
" chunk=eval_config.chunk)\n",
"train_step = functools.partial(\n",
" training.train_step,\n",
" model,\n",
" elastic_reduce_method=train_config.elastic_reduce_method,\n",
" elastic_loss_type=train_config.elastic_loss_type,\n",
" use_elastic_loss=train_config.use_elastic_loss,\n",
" use_background_loss=train_config.use_background_loss,\n",
" use_warp_reg_loss=train_config.use_warp_reg_loss,\n",
" use_hyper_reg_loss=train_config.use_hyper_reg_loss,\n",
")\n",
"ptrain_step = jax.pmap(\n",
" train_step,\n",
" axis_name='batch',\n",
" devices=devices,\n",
" # rng_key, state, batch, scalar_params.\n",
" in_axes=(0, 0, 0, None),\n",
" # Treat use_elastic_loss as compile-time static.\n",
" donate_argnums=(2,), # Donate the 'batch' argument.\n",
")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 1000
},
"id": "vbc7cMr5aR_1",
"outputId": "d35e110d-7dbc-41ca-acae-4f81d0a5af22"
},
"outputs": [],
"source": [
"# @title Train!\n",
"# @markdown This runs the training loop!\n",
"\n",
"import mediapy\n",
"from hypernerf import utils\n",
"from hypernerf import visualization as viz\n",
"\n",
"\n",
"print_every_n_iterations = 100 # @param{type:'number'}\n",
"visualize_results_every_n_iterations = 500 # @param{type:'number'}\n",
"save_checkpoint_every_n_iterations = 1000 # @param{type:'number'}\n",
"\n",
"\n",
"logging.info('Starting training')\n",
"rng = rng + jax.process_index() # Make random seed separate across hosts.\n",
"keys = random.split(rng, len(devices))\n",
"time_tracker = utils.TimeTracker()\n",
"time_tracker.tic('data', 'total')\n",
"\n",
"for step, batch in zip(range(step, train_config.max_steps + 1), train_iter):\n",
" time_tracker.toc('data')\n",
" scalar_params = scalar_params.replace(\n",
" learning_rate=learning_rate_sched(step),\n",
" elastic_loss_weight=elastic_loss_weight_sched(step))\n",
" # pytype: enable=attribute-error\n",
" nerf_alpha = jax_utils.replicate(nerf_alpha_sched(step), devices)\n",
" warp_alpha = jax_utils.replicate(warp_alpha_sched(step), devices)\n",
" hyper_alpha = jax_utils.replicate(hyper_alpha_sched(step), devices)\n",
" hyper_sheet_alpha = jax_utils.replicate(\n",
" hyper_sheet_alpha_sched(step), devices)\n",
" state = state.replace(nerf_alpha=nerf_alpha,\n",
" warp_alpha=warp_alpha,\n",
" hyper_alpha=hyper_alpha,\n",
" hyper_sheet_alpha=hyper_sheet_alpha)\n",
"\n",
" with time_tracker.record_time('train_step'):\n",
" state, stats, keys, _ = ptrain_step(keys, state, batch, scalar_params)\n",
" time_tracker.toc('total')\n",
"\n",
" if step % print_every_n_iterations == 0:\n",
" logging.info(\n",
" 'step=%d, warp_alpha=%.04f, hyper_alpha=%.04f, hyper_sheet_alpha=%.04f, %s',\n",
" step, \n",
" warp_alpha_sched(step), \n",
" hyper_alpha_sched(step), \n",
" hyper_sheet_alpha_sched(step), \n",
" time_tracker.summary_str('last'))\n",
" coarse_metrics_str = ', '.join(\n",
" [f'{k}={v.mean():.04f}' for k, v in stats['coarse'].items()])\n",
" fine_metrics_str = ', '.join(\n",
" [f'{k}={v.mean():.04f}' for k, v in stats['fine'].items()])\n",
" logging.info('\\tcoarse metrics: %s', coarse_metrics_str)\n",
" if 'fine' in stats:\n",
" logging.info('\\tfine metrics: %s', fine_metrics_str)\n",
" \n",
" if step % visualize_results_every_n_iterations == 0:\n",
" print(f'[step={step}] Training set visualization')\n",
" eval_batch = next(train_eval_iter)\n",
" render = render_fn(state, eval_batch, rng=rng)\n",
" rgb = render['rgb']\n",
" acc = render['acc']\n",
" depth_exp = render['depth']\n",
" depth_med = render['med_depth']\n",
" rgb_target = eval_batch['rgb']\n",
" depth_med_viz = viz.colorize(depth_med, cmin=datasource.near, cmax=datasource.far)\n",
" mediapy.show_images([rgb_target, rgb, depth_med_viz],\n",
" titles=['GT RGB', 'Pred RGB', 'Pred Depth'])\n",
"\n",
" print(f'[step={step}] Validation set visualization')\n",
" eval_batch = next(val_eval_iter)\n",
" render = render_fn(state, eval_batch, rng=rng)\n",
" rgb = render['rgb']\n",
" acc = render['acc']\n",
" depth_exp = render['depth']\n",
" depth_med = render['med_depth']\n",
" rgb_target = eval_batch['rgb']\n",
" depth_med_viz = viz.colorize(depth_med, cmin=datasource.near, cmax=datasource.far)\n",
" mediapy.show_images([rgb_target, rgb, depth_med_viz],\n",
" titles=['GT RGB', 'Pred RGB', 'Pred Depth'])\n",
"\n",
" if step % save_checkpoint_every_n_iterations == 0:\n",
" training.save_checkpoint(checkpoint_dir, state)\n",
"\n",
" time_tracker.tic('data', 'total')\n"
]
}
],
"metadata": {
"accelerator": "GPU",
"colab": {
"gpuType": "V100",
"machine_shape": "hm",
"provenance": []
},
"gpuClass": "standard",
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.10.10"
}
},
"nbformat": 4,
"nbformat_minor": 1
}
|