Nerfstudio: A collaboration friendly studio for NeRFs

This colab shows how to train and view NeRFs from Nerfstudio both on pre-made datasets or from your own videos/images.

Credit to NeX for Google Colab format.

Frequently Asked Questions

Downloading custom data is stalling (no output):
- This is a bug in Colab. The data is processing, but may take a while to complete. You will know processing completed if data/nerfstudio/custom_data/transforms.json exists.
Training is not showing progress:
- The lack of output is a bug in Colab. You can see the training progress from the viewer.
Viewer Quality is bad / Low resolution:
- This may be because more GPU is being used on training that rendering the viewer. Try pausing training or decreasing training utilization.
Other problems?
- Feel free to create an issue on our GitHub repo.

In [ ]:

#@title # Install Conda (requires runtime restart) { vertical-output: true, display-mode: "form" }

!pip install -q condacolab
import condacolab
condacolab.install()

⏬ Downloading https://github.com/jaimergp/miniforge/releases/latest/download/Mambaforge-colab-Linux-x86_64.sh...
📦 Installing...
📌 Adjusting configuration...
🩹 Patching environment...
⏲ Done in 0:00:23
🔁 Restarting kernel...

In [ ]:

#@title # Install Nerfstudio and Dependencies (~10 min) { vertical-output: true, display-mode: "form" }

%cd /content/
!pip install --upgrade pip
!pip install torch==1.12.1+cu113 torchvision==0.13.1+cu113 -f https://download.pytorch.org/whl/torch_stable.html

# Installing TinyCuda
%cd /content/
!gdown "https://drive.google.com/u/1/uc?id=1q8fuc-Mqiev5GTBTRA5UPgCaQDzuqKqj" 
!pip install tinycudann-1.6-cp37-cp37m-linux_x86_64.whl

# Installing COLMAP
%cd /content/
!conda install -c conda-forge colmap

# Install nerfstudio
%cd /content/
!pip install nerfstudio

In [ ]:

#@markdown <h1> Downloading Data</h1>
#@markdown <h3>Pick the preset scene or upload your own images/video</h3>
import os
from google.colab import files
from IPython.core.display import display, HTML

scene = '\uD83D\uDDBC poster' #@param ['🖼 poster', '🚜 dozer', '🌄 desolation', '📤 upload your images' , '🎥 upload your own video']
scene = ' '.join(scene.split(' ')[1:])

if scene not in ['upload your images', 'upload your own video']:
    %cd /content/
    !ns-download-data --dataset=nerfstudio --capture=$scene
else:
    display(HTML('<h3>Select your custom data</h3>'))
    display(HTML('<p/>You can select multiple images by pressing ctrl, cmd or shift and click.<p>'))
    display(HTML('<p/>Note: This may take time, especially on hires inputs, so we recommend to download dataset after creation.<p>'))
    !mkdir -p /content/data/nerfstudio/custom_data
    if scene == 'upload your images':
        !mkdir -p /content/data/nerfstudio/custom_data/raw_images
        %cd /content/data/nerfstudio/custom_data/raw_images
        uploaded = files.upload()
        dir = os.getcwd()
    else:
        %cd /content/data/nerfstudio/custom_data/
        uploaded = files.upload()
        dir = os.getcwd()
    preupload_datasets = [os.path.join(dir, f) for f in uploaded.keys()]
    del uploaded
    %cd /content/

    if scene == 'upload your images':
        !ns-process-data images --data /content/data/nerfstudio/custom_data/raw_images --output-dir /content/data/nerfstudio/custom_data/
    else:
        video_path = preupload_datasets[0]
        !ns-process-data video --data $video_path --output-dir /content/data/nerfstudio/custom_data/

    scene = "custom_data"

/content
Downloading...
From: https://drive.google.com/uc?id=1dmjWGXlJnUxwosN6MVooCDQe970PkD-1
To: /content/data/nerfstudio/poster.zip
100% 750M/750M [00:03<00:00, 218MB/s]
[0m

In [ ]:

#@title # Set Up Viewer

%cd /content

# Install localtunnel
# We are using localtunnel https://github.com/localtunnel/localtunnel but ngrok could also be used
!npm install -g localtunnel

# Tunnel port 7007, the default for
!rm url.txt 2> /dev/null
get_ipython().system_raw('lt --port 7007 >> url.txt 2>&1 &')

/content
[K[?25h/tools/node/bin/lt -> /tools/node/lib/node_modules/localtunnel/bin/lt.js
+ localtunnel@2.0.2
added 22 packages from 22 contributors in 1.536s

In [ ]:

#@title # Start Viewer

with open('url.txt') as f:
  lines = f.readlines()
websocket_url = lines[0].split(": ")[1].strip().replace("https", "wss")
# from nerfstudio.utils.io import load_from_json
# from pathlib import Path
# json_filename = "nerfstudio/nerfstudio/viewer/app/package.json"
# version = load_from_json(Path(json_filename))["version"]
url = f"https://viewer.nerf.studio/?websocket_url={websocket_url}"
print(url)
print("You may need to click Refresh Page after you start training!")
from IPython import display
display.IFrame(src=url, height=800, width="100%")

https://viewer.nerf.studio/?websocket_url=wss://good-ways-ask-34-83-255-230.loca.lt
You may need to click Refresh Page after you start training!

In [ ]:

#@title # Start Training { vertical-output: true }

%cd /content
!ns-train nerfacto --viewer.websocket-port 7007 nerfstudio-data --data data/nerfstudio/$scene --downscale-factor 4

In [ ]:

#@title # Render Video { vertical-output: true }
#@markdown <h3>Export the camera path from within the viewer, then run this cell.</h3>
#@markdown <h5>The rendered video should be at renders/output.mp4!</h5>


base_dir = "/content/outputs/data-nerfstudio-" + scene + "/nerfacto/"
training_run_dir = base_dir + os.listdir(base_dir)[0]

from IPython.core.display import display, HTML
display(HTML('<h3>Upload the camera path JSON.</h3>'))
%cd $training_run_dir
uploaded = files.upload()
uploaded_camera_path_filename = list(uploaded.keys())[0]

config_filename = training_run_dir + "/config.yml"
camera_path_filename = training_run_dir + "/" + uploaded_camera_path_filename
camera_path_filename = camera_path_filename.replace(" ", "\\ ").replace("(", "\\(").replace(")", "\\)")

%cd /content/
!ns-render --load-config $config_filename --traj filename --camera-path-filename $camera_path_filename --output-path renders/output.mp4

[2;36m[19:48:48][0m[2;36m [0mSkipping [1;36m0[0m files in dataset split train.                                          ]8;id=527413;file:///content/nerfstudio/nerfstudio/data/dataparsers/nerfstudio_dataparser.py\[2mnerfstudio_dataparser.py[0m]8;;\[2m:[0m]8;id=243595;file:///content/nerfstudio/nerfstudio/data/dataparsers/nerfstudio_dataparser.py#91\[2m91[0m]8;;\
[2;36m          [0m[2;36m [0mSkipping [1;36m0[0m files in dataset split test.                                           ]8;id=109270;file:///content/nerfstudio/nerfstudio/data/dataparsers/nerfstudio_dataparser.py\[2mnerfstudio_dataparser.py[0m]8;;\[2m:[0m]8;id=464675;file:///content/nerfstudio/nerfstudio/data/dataparsers/nerfstudio_dataparser.py#91\[2m91[0m]8;;\
[2KLoading data batch [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [35m100%[0m [36m0:00:00[0m
[?25h/usr/local/lib/python3.7/site-packages/torch/utils/data/dataloader.py:566: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.
  cpuset_checked))
[2KLoading data batch [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [35m100%[0m [36m0:00:00[0m
[?25h/usr/local/lib/python3.7/site-packages/torchvision/models/_utils.py:209: UserWarning: The parameter 'pretrained' is deprecated since 0.13 and will be removed in 0.15, please use 'weights' instead.
  f"The parameter '{pretrained_param}' is deprecated since 0.13 and will be removed in 0.15, "
/usr/local/lib/python3.7/site-packages/torchvision/models/_utils.py:223: UserWarning: Arguments other than a weight enum or `None` for 'weights' are deprecated since 0.13 and will be removed in 0.15. The current behavior is equivalent to passing `weights=AlexNet_Weights.IMAGENET1K_V1`. You can also use `weights=AlexNet_Weights.DEFAULT` to get the most up-to-date weights.
  warnings.warn(msg)
Loading latest checkpoint from load_dir
✅ Done loading checkpoint from 
outputs/data-nerfstudio-poster/nerfacto/[1;36m2022[0m-[1;36m10[0m-29_192844/nerfstudio_models/step-[1;36m000014000.[0mckpt
[1;32mCreating trajectory video[0m
[2K🎥 Rendering 🎥 [90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━[0m [35m100%[0m [31m0.14 fps[0m [33m11:47[0m
[2K[32m(  ●   )[0m [33mSaving video[0m
[1A[2K[92m────────────────────────────────────────────── [0m[32m 🎉 🎉 🎉 Success 🎉 🎉 🎉[0m[92m ──────────────────────────────────────────────[0m
                                           [32mSaved video to renders/output.mp4[0m                                            
[0m

17 KiB Raw Permalink Blame History

Nerfstudio: A collaboration friendly studio for NeRFs

Frequently Asked Questions

17 KiB

Raw Permalink Blame History