nerfstudio/notebooks/visualize_dataset.ipynb

Notebook to visualize a dataset

In [ ]:

from notebook_imports import *

import torch
from torch.utils.data import DataLoader

from radiance.nerf.dataset.image_dataset import ImageDataset, collate_batch
from radiance.nerf.dataset.collate import CollateIterDataset, collate_batch_size_one
from radiance.nerf.dataset.utils import get_dataset_inputs
from radiance.nerf.field_modules.ray_generator import RayGenerator
from radiance.nerf.field_modules.scene_bounds_collider import SceneBoundsCollider
from radiance.structures.rays import RayBundle
from radiance.utils.io import get_absolute_path
from radiance.viewer.plotly import get_line_segments_from_lines
from radiance.structures.cameras import get_camera_model


from hydra import compose, initialize
from omegaconf import open_dict
import pprint
from tqdm import tqdm
import random

In [ ]:

import goat

vis = goat.vis_utils.get_vis()

In [ ]:

goat.vis_utils.show_box_test(vis)

In [ ]:

with initialize(config_path="../configs"):
    cfg = compose(config_name="default.yaml")
    # cfg = compose(config_name="default.yaml", overrides=["data/dataset=friends_TBBT-big_living_room"])

In [ ]:

dataset_inputs = get_dataset_inputs(**cfg.data.dataset, split="train")

In [ ]:

# import goat.threedvis.meshcat.geometry as g
# import numpy as np
# point_cloud = dataset_inputs.point_cloud
# size = 0.1
# vis["point_cloud"].set_object(
#     g.PointCloud(position=np.transpose(np.array(point_cloud.xyz)),
#                     color=np.transpose(np.array(point_cloud.rgb)) / 255.0, size=size)
# )

In [ ]:

# show dataset inputs with three.js
# goat.vis_utils.show_box_test(vis)
num_images_to_show = 200
for idx in tqdm(list(range(len(dataset_inputs.image_filenames)))[:num_images_to_show]):
    image = imageio.imread(dataset_inputs.image_filenames[idx])[:, :, :3]

    num_intrinsics_params = len(dataset_inputs.intrinsics[idx])
    camera_class = get_camera_model(num_intrinsics_params)
    camera_to_world = dataset_inputs.camera_to_world[idx]
    camera = camera_class(*dataset_inputs.intrinsics[idx].tolist(), camera_to_world=camera_to_world)
    K = camera.get_intrinsics_matrix()

    name = "train/{:06d}".format(idx)
    # height = 0.5
    displayed_focal_length = 0.5
    goat.vis_utils.draw_camera_frustum(
        vis,
        name=name,
        image=image,
        K=K.numpy(),
        pose=camera_to_world.double().numpy(),
        height=3.0,
        realistic=False,
        displayed_focal_length=0.0,
    )

In [ ]:

image_dataset = ImageDataset(
    image_filenames=dataset_inputs.image_filenames, downscale_factor=dataset_inputs.downscale_factor
)
iter_dataset = CollateIterDataset(
    image_dataset,
    collate_fn=lambda batch_list: collate_batch(
        batch_list, cfg.data.dataloader.num_rays_per_batch, keep_full_image=True
    ),
    num_samples_to_collate=cfg.data.dataloader.num_images_to_sample_from,
    num_times_to_repeat=cfg.data.dataloader.num_times_to_repeat_images,
)
dataloader = DataLoader(
    iter_dataset,
    batch_size=1,
    num_workers=cfg.data.dataloader.num_workers,
    collate_fn=collate_batch_size_one,
    pin_memory=True,
)
dataloader_iter = iter(dataloader)

ray_generator = RayGenerator(dataset_inputs.intrinsics, dataset_inputs.camera_to_world)

num_batches = 10
for _ in tqdm(range(num_batches)):
    batch = next(dataloader_iter)
    ray_bundle = ray_generator.forward(batch["indices"])
    break

In [ ]:

media.show_image(image_dataset.get_image(10))

In [ ]:

def show_batch(batch):
    # set the color of the sampled rays
    c, y, x = [i.flatten() for i in torch.split(batch["local_indices"], 1, dim=-1)]
    batch["image"][c, y, x] = 1.0

    # batch["image"] is num_images, h, w, 3
    images = torch.split(batch["image"], 1, dim=0)
    image_list = [image[0] for image in images]
    image = torch.cat(image_list, dim=1)  # cat along the width dimension

    # the white pixels are rays
    media.show_image((image * 255).to(torch.uint8))


def sample_and_show_batch():
    batch = next(iter(dataloader))
    show_batch(batch)
    return batch

In [ ]:

batch = sample_and_show_batch()

In [ ]:

ray_indices = batch["indices"]
ray_bundle = ray_generator(ray_indices)

In [ ]:

skip = 1
size = 8
data = []
data += [
    go.Scatter3d(
        x=ray_generator.camera_to_world[::skip, 0, 3],
        y=ray_generator.camera_to_world[::skip, 1, 3],
        z=ray_generator.camera_to_world[::skip, 2, 3],
        mode="markers",
        name="origins",
        marker=dict(color="rgba(0, 0, 0, 1)", size=size),
    )
]

In [ ]:

scene_bounds_collider = SceneBoundsCollider(dataset_inputs.scene_bounds)
intersected_ray_bundle = scene_bounds_collider(ray_bundle)

In [ ]:

# intersected_ray_bundle.nears

In [ ]:

# near_plane = cfg.graph.network.near_plane
# far_plane = cfg.graph.network.far_plane
lines = torch.stack(
    [
        intersected_ray_bundle.origins + intersected_ray_bundle.directions * intersected_ray_bundle.nears,
        intersected_ray_bundle.origins + intersected_ray_bundle.directions * intersected_ray_bundle.fars,
    ],
    dim=1,
).tolist()  # (num_rays, 2, 3)
lines = torch.tensor(random.sample(lines, k=100))

In [ ]:

data += get_line_segments_from_lines(lines)

In [ ]:

layout = go.Layout(
    autosize=False,
    width=1000,
    height=1000,
    margin=go.layout.Margin(l=50, r=50, b=100, t=100, pad=4),
    scene=go.layout.Scene(
        aspectmode="data",
        camera=dict(up=dict(x=0, y=0, z=1), center=dict(x=0, y=0, z=0), eye=dict(x=1.25, y=1.25, z=1.25)),
    ),
)
fig = go.Figure(data=data, layout=layout)
fig.show()

In [ ]:

In [ ]:

fig.write_image("temp.png")
plotly_image = imageio.imread("temp.png")

In [ ]:

In [ ]: