README.md

Running Caffe Inception_v2 SSD Model:

Data Preparation

Download VOC2007 dataset.

# Download the data.
wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtest_06-Nov-2007.tar
# Extract the data.
tar -xvf VOCtest_06-Nov-2007.tar
#Generate ground truth file 
python generate_gt_file.py

📌 NOTE: VOC dataset contains 21 classes. But this model is trained with 19 classes (removed diningtable and train). If your model is having 21 classes, comment 46 line in generate_gt_file.py

The format of calib.txt used in calibration phase of decent_q is as follow:

#image_name fake_label_number  
000001.jpg 1
000002.jpg 1
000003.jpg 1
000004.jpg 1
000006.jpg 1

📌 NOTE: The label number is not actually used in calibration and arbitrary label number can be used.

# Get the necessary models
python $MLSUITE_ROOT/examples/caffe/getModels.py

# Setup ml-suite Environment Variables
source $MLSUITE_ROOT/overlaybins/setup.sh

Prepare a model for inference

** MUST BE DONE FIRST ** To run a Caffe model on the FPGA, it needs to be quantized, compiled, and a new graph needs to be generated. The new graph is similar to the original, with the FPGA subgraph removed, and replaced with a custom Python layer.

python run_ssd.py --prototxt /opt/models/caffe/inception_v2_ssd/inception_v2_ssd_train.prototxt --caffemodel /opt/models/caffe/inception_v2_ssd/inception_v2_ssd.caffemodel --prepare

Run Inference on entire dataset and caluculate mAP

python run_ssd.py --prototxt xfdnn_auto_cut_deploy.prototxt --caffemodel quantize_results/deploy.caffemodel --labelmap_file labelmap_voc_19c.prototxt --test_image_root ./VOCdevkit/VOC2007/JPEGImages/ --image_list_file ./VOCdevkit/VOC2007/ImageSets/Main/test.txt --gt_file voc07_gt_file_19c.txt --validate

Run Inference for a single image

python run_ssd.py --prototxt xfdnn_auto_cut_deploy.prototxt --caffemodel quantize_results/deploy.caffemodel --labelmap_file labelmap_voc_19c.prototxt --image ./test_pic/000022.jpg