This page lists a number of example SegNet models in the SegNet Model Zoo. NOTE: all Bayesian SegNet models can be tested as SegNet models (for example by using the webcam demo) by removing the line sample_weights_test: true on all Dropout layers, and setting batch size of 1.
This example model is used in the SegNet webdemo [http://mi.eng.cam.ac.uk/projects/segnet/]. It is trained to classify road scenes into 12 classes.
Model file: segnet_model_driving_webdemo.prototxt
Weights can be downloaded from this link: [http://mi.eng.cam.ac.uk/~agk34/resources/SegNet/segnet_weights_driving_webdemo.caffemodel]
These models have been trained for road scene understanding using the CamVid dataset.
- Segnet Basic model file:
segnet_basic_camvid.prototxtweights: [http://mi.eng.cam.ac.uk/~agk34/resources/SegNet/segnet_basic_camvid.caffemodel] - Bayesian Segnet model file:
bayesian_segnet_camvid.prototxtweights: [http://mi.eng.cam.ac.uk/~agk34/resources/SegNet/bayesian_segnet_camvid.caffemodel] - Bayesian Segnet Basic model file:
bayesian_segnet_basic_camvid.prototxtweights: [http://mi.eng.cam.ac.uk/~agk34/resources/SegNet/bayesian_segnet_basic_camvid.caffemodel]
These models have been trained for indoor scene understanding using the SUN RGB-D dataset.
- Segnet model file:
segnet_sun.prototxtweights: [http://mi.eng.cam.ac.uk/~agk34/resources/SegNet/segnet_sun.caffemodel] - Segnet model file:
bayesian_segnet_sun.prototxtweights: [http://mi.eng.cam.ac.uk/~agk34/resources/SegNet/segnet_sun.caffemodel]
The model definition file used for training can be found here train_segnet_sun.prototxt
We have also trained a model for a 224x224 pixel input:
- Segnet low resolution model file:
segnet_sun_low_resolution.prototxtweights: [http://mi.eng.cam.ac.uk/~agk34/resources/SegNet/segnet_sun_low_resolution.caffemodel]
These models have been trained on the Pascal VOC 2012 dataset .
- Segnet model file:
segnet_pascal.prototxtweights: [http://mi.eng.cam.ac.uk/~agk34/resources/SegNet/segnet_pascal.caffemodel] - Bayesian Segnet model file:
bayesian_segnet_pascal.prototxtweights: [http://mi.eng.cam.ac.uk/~agk34/resources/SegNet/segnet_pascal.caffemodel]
This model is based on the Dropout enc-dec variant and is designed for a 224x224 pixel input.
This model finetuned the webdemo weights using the CityScapes dataset (11 class version).
- 11 Class CityScapes Model (trained by Timo Sämann, Aschaffenburg University of Applied Sciences) model file:
segnet_model_driving_webdemo.prototxtweights: [http://mi.eng.cam.ac.uk/~agk34/resources/SegNet/segnet_iter_30000_timo.caffemodel]
These models are released under a creative commons license which allows for personal and research use only. For a commercial license please contact the authors. You can view a license summary here: http://creativecommons.org/licenses/by-nc/4.0/
Alex Kendall [email protected] Cambridge University