#This is a research project on the use of a single regression convolutional neural network to detect small vehicle targets in aerial photography. The corresponding paper is under review. If it is published, we will update it here.
- Please refer to ssd installation method to compile ssd-caffe.
The website is https://github.com/weiliu89/caffe/tree/ssd
-
Unzip the penlu.tar.gz file from this project. Then place the file in the corresponding location in ssd-caffe. Then recompile the new caffe.
-
Please download the data and training file LSFSCD_train_test.tar.gz at https://pan.baidu.com/s/1w-Mk_2cSrVSmIf58pvqXOQ (password: cb8t)
-
Extract the LSFSCD_train_test.tar.gz file from this project. This includes the data and network structure.
-
After the third step of decompression, the corresponding path in each file is edited according to the actual situation of the reader. Execute the file train_val/VGG_VEDAI512_512x512.sh to train the model.
-
After the training model is over, use LSFSCD_detect.py to test to evaluate our model.
-
It should be noted that the training model requires fine tuning using VGG_ILSVRC_16_layers_fc_reduced.caffemodel.