pubgeo / dfc2019 Goto Github PK

2019 IEEE GRSS Data Fusion Contest data, baselines, and metrics

License: MIT License

Python 89.70% Dockerfile 0.14% Shell 0.82% C++ 5.56% Cuda 3.78%

dfc2019's Introduction

SHR3D

SHR3D classifies a 3D point cloud to generate simple geospatial products including a Digital Surface Model (DSM); a Digital Terrain Model (DTM); a classification image labeling ground, buildings, and trees; and a simplified building outlines vector product.

SHR3D Usage

./shr3d <Input File (LAS|TIF)> DH=<Horizontal uncertainty(m)> DZ=<Vertical uncertainty(m)> AGL=<Min building height(m)> <Options>
Options:
    AREA=<Min building area(m^2)>   Buildings smaller than this size will not be labeled (default: 50)
    EGM96                           Set this flag to write vertical datum
    BOUNDS=<MINX,MAXX,MINY,MAXY>    Set to define image bounds
    DTM=<DTM File (TIF)>            Path to optional DTM file
    GND_LABEL=<LABEL>               Set to the value of the point cloud ground classification label (typically 2) if the point cloud has already been partially classified.

Example:

./shr3d shr3dMe.las DH=0.5 DZ=0.5 AGL=2 AREA=50

This will produce multiple files as a result:

Raster files (*.tif):
- Classification image- 'Colored' by classification for each pixel: buildings, ground, and trees
- Building image- Binary mask of all detected buildings
- DSM- Digital surface model (reflective surface)
- DTM- Digital terrain model (bare earth, or ground)
- INT- Intensity image
Vector files (shapefiles: *.shp, *.prj, *.shx, *.dbf):
- Building outlines- Simplified polygons of detected building perimeters

ALIGN3D

ALIGN3D estimates and applies a transform to align 3D point clouds. This algorithm was developed for use with airborne lidar, multiple view satellite imagery, and synthetic aperture radar derived point clouds.

ALIGN3D Usage

./align3d <Reference point cloud(LAS)> <Target point cloud(LAS)> maxdz=<Maximum local z difference(m)> gsd=<Ground sample distance(m)> maxt=<Maximum XY translation for search(m)>

Example:

./align3d reference.las target.las maxt=10.0 gsd=1.0

This will produce multiple output files:

Aligned.las- A point cloud representation of the target, aligned to the reference
Aligned.tif- A DSM of the target, aligned to the reference
Offsets.txt- A list of the xyz offsets to translate the target to align with the reference

dfc2019's People

Contributors

Stargazers

Watchers

dfc2019's Issues

about track4 pointnet2 training data

Hello,
I retrained pointnet2 with provided codes on 99 point clouds which are produced by create_train_dataset.py and tested on 11 validation point clouds which are produced by create_train_dataset.py. But it is 6% lower than provided trained model. Does the baseline use a different split?

Thank you very much!

An error in Track4

the interface.py needs "import provider", but in the docker image, the python version is 3.5 which is lower than the minimum requirement(python >=3.6) of the "provider". Does anyone meet the same problem?

Including RGB in the .las

@pubgeo I am using this version of Pointnet++, my dataset has yxzRGB values in it. But when i tried to feed in the RGB values to the .las using the laspy somehow it is not happening.
Can you help please

Track 1-3 Metrics

Memory error preventing 3d component of track1-3 metrics from being released. Updated metrics are on the way.

ValueError('all input arrays must have the same shape') in create_train_dataset.py

I am trying to run pointnet2 on my own data set, which has got a '_PC3.txt' file with xyzRGB values and a corresponding '_CLS.txt' file. I am getting a error like this when I ran create_train_dataset.py

But i checked the shape of CLS and PC3 file, both are same. Is it a problem because of multi-threading? If so how can I rectify it?
I tried in the LiNUX environment as well. But it gave the same error.
Please help

Get lfs out of quota

fetch: Fetching reference refs/heads/master
batch response: This repository is over its data quota. Purchase more data packs to restore access.
error: failed to fetch some objects from 'https://github.com/pubgeo/dfc2019.git/info/lfs'

Will the baseline results be released?

Will you plan to provide the baseline results on val set?

From where can I download the whole datasets please?

Hello there, it's been several days since the release of the training and validation dataset. Yet I couldn't pull these data to the local till now, though I had registered on the official website to make the links visible, all as guided. For the first few days, the links contained in the BT files seemed invalid, as the resources couldn't be connected. And today found the links to the BT files and the net disks on the official page all gone.

As is the case, could you please show me another way to download the dataset? Sorry for any possible trouble on reading caused by languages issues as my English is not so good.

Downloading the track4 data by torrent is too slow. The downloading speed is less than 1kb/s.

Is it possibile to submit predictions to the leaderboard of the DFC2019 challenge?

Hi, @pubgeo, I want to submit predictions to the leaderboard of the DFC2019 challenge, but it said "Your request to participate in this challenge has been received and a decision is pending." in the website https://competitions.codalab.org/competitions/20212#participate. Do you know how long it will take? Or whether the leaderboard has already closed?

Weight files download

When I git lfs clone this repo, the result is:
This repository is over its data quota. Account responsible for
LFS bandwidth should purchase more data packs to restore access.

Is there any other way to download the weight file? Thank you!

AttributeError: module 'tensorflow._api.v2.image' has no attribute 'resize_bilinear'

After setting up anaconda with conda env create --name dfc2019 --file=dev-gpu.yml
and running

conda activate dfc2019
cd track3
python ./mvs/test-mvs.py

I receive the following error message

Traceback (most recent call last):
  File "./mvs/test-mvs.py", line 693, in <module>
    predictor.build_seg_model(seg_weights_file)
  File "./mvs/test-mvs.py", line 103, in build_seg_model
    self.seg_model = build_icnet(self.height, self.width, self.bands, self.num_categories + 1,
  File "/mnt/Data-512GB/libraries_ml_geo/dfc2019/dfc2019/track3/mvs/model_icnet.py", line 52, in build_icnet
    y = Lambda(lambda x: tf.image.resize_bilinear(x, size=(int(x.shape[1])//2, int(x.shape[2])//2)), name='data_sub2')(inp)
  File "/home/sebastian/miniconda3/envs/dfc2019/lib/python3.8/site-packages/tensorflow/python/keras/engine/base_layer.py", line 922, in __call__
    outputs = call_fn(cast_inputs, *args, **kwargs)
  File "/home/sebastian/miniconda3/envs/dfc2019/lib/python3.8/site-packages/tensorflow/python/keras/layers/core.py", line 888, in call
    result = self.function(inputs, **kwargs)
  File "/mnt/Data-512GB/libraries_ml_geo/dfc2019/dfc2019/track3/mvs/model_icnet.py", line 52, in <lambda>
    y = Lambda(lambda x: tf.image.resize_bilinear(x, size=(int(x.shape[1])//2, int(x.shape[2])//2)), name='data_sub2')(inp)
AttributeError: module 'tensorflow._api.v2.image' has no attribute 'resize_bilinear'

Unrectified images and corrected RPC for Track2

Where can I get access to unrectified images and corrected RPC or affine camera models?

The following is mentioned on page4 of the publication "Metadata such as RPC, epipolar rectifying homographies, and collection dates are retained for each stereo pair" I am not able to find RPC parameters in METADATA.json files or any of the .tif files.

0 peers for torrent

I tried some of the provided torrents and all seem to be broken.

pubgeo / dfc2019 Goto Github PK

dfc2019's Introduction

SHR3D

SHR3D Usage

Example:

ALIGN3D

ALIGN3D Usage

Example:

dfc2019's People

Contributors

Stargazers

Watchers

Forkers

dfc2019's Issues

Recommend Projects

Recommend Topics

Recommend Org