Welcome to dacl.ai!
dacl.ai is a platform for damage recognition enthusiasts who want to compare and share their
models with the dacl community
Coupled to dacl.ai is bikit which
was introduced in the paper: "Building Inspection Toolkit: Unified Evaluation and Strong Baselines for Damage Recognition".
bikit is a simple-to-use Python API for machine learning projects in the field of damage detection for built structures, currenlty with a focus on reinforced concrete bridges.
All models displayed inside dacl.ai's leaderboard are also accessible via bikit.
At the moment, dacl.ai tackles bikit's datasets and models for multi-target classification.
Other data such as single-target, object detection and semantic segmentation datasets will follow.
You can compare models by using the leaderboard below showing the most important metrics inside of one table and two charts.
If you want to contribute and share your results you can do so by submitting results here.
Leaderboard
The leaderboard shows the results on test data of all models that were submitted to dacl.ai listed according to the EMR. Models marked with a medal represent the leader in the corresponding dataset. You can select which models, depending on their training data, you want to display by marking the checkbox above the leaderboard.
Loading...
EMR
This chart displays the progress of the
submitted models or rather their Exact Match Ratio (EMR) over their released date. EMR is the most
important metric for multi-target classification. Further information regarding the models and the metrics may
be found here.
For filtering the results, the labels under each chart can be toggled.
F1-score
Recall by class
Get datasets and models
Install bikit to download datasets and pretrained models.
Improve a model or train one from scratch
Improve a pretrained model or train a model from scratch with an architecture of your choice.