Geometry method

results

environments

• pytorch1.0
• python3.5

Prepare

# compile lib
cd lib
python setup.py build develop
cd -

Training

download pretrained model(res50, res18_pruned) from smb://deepmotion1/public/liangxiong first

CUDA_VISIBLE_DEVICES=1 python trainval_net.py --cuda \
    --net mono_3d_final_plus \
    --out_path /data/object/liangxiong/coco_pretrained_normalized_refine \
    --config configs/coco_mono_3d_config.json

you need to check run.sh first then run sh run.sh

Testing

mkdir results/data -p

CUDA_VISIBLE_DEVICES=0 python test_net.py --cuda \ 
    --net mono_3d_final_plus \
    --load_dir /data/object/liangxiong/coco_pretrained_normalized_refine \
    --checkpoint 4912 \
    --checkepoch 49
    #    --img_dir /data/dm202_3w/left_img \
    #    --calib_file ./000004.txt

results will saved to results/data

Visualization

mkdir results/fv
python vis_all_3d.py