SqueezeDet for gesture
Forked from SqueezeDet and modified for gesture images in Pascol VOC format.
Original paper: SqueezeDet: Unified, Small, Low Power Fully Convolutional Neural Networks for Real-Time Object Detection for Autonomous Driving. Bichen Wu, Forrest Iandola, Peter H. Jin, Kurt Keutzer (UC Berkeley & DeepScale)
CNN pretrained Model:
Training/Validation:
Prepare the Dataset in Psacol Voc style:
$Path_to_data/VOC2007/ |->Annotations/ | |-> 00****.xml
L-> ...
|->JPEGImages/ | L-> 00****.jpg
L->ImageSets/Main |-> test.txt |-> train.txt |-> trainval.txt
L-> val.txt
Now we can start training. Training script can be found in $SQDT_ROOT/scripts/train.sh
cd $SQDT_ROOT/
./scripts/train.sh squeezeDet
At the same time, you can launch evaluation by
cd $SQDT_ROOT/
./scripts/eval_train.sh
./scripts/eval_val.sh
Finally, to monitor training and evaluation process, you can use tensorboard by
tensorboard --logdir=$LOG_DIR
Demo