PIDNet Model for City-Scapes Dataset

Requirements

Install all requirements in requirements.txt file:

pip3 install -r requirements.txt

Train model

!! Before runing this command, you must change mode of this bash file by :

chmod +x run_train.sh

Training model by running this command

./run_train.sh

Evaluation

!! Before runing this command, you must change mode of this bash file by :

chmod +x run_test.sh

Evaluate model by running this command

./run_test.sh

Dataset Architecture


data
    |
    |CityScapes
                |----gtFine
                |        |-- test
                |        |-- train
                |        |-- val
                |    
                |---- leftImg8bit
                         |-- test
                         |-- train
                         |-- val

Change model PIDNet to PIDNet + DCN or PIDNet + MDCN

If you want to change model from PIDNet to PIDNet + DCN, you can change this line in models/model_utils.py

def __init__(self, inplanes, planes, stride=1, downsample=None, no_relu=False, use_dcn=False, modulated=True):

If you want to change to PIDNet + DCN, seting use_dcn=True and modulated=False

If you want to change to PIDNet + MDCN, setting use_dcn=True and modulated=True

Checkpoint

Checkpoint models are located in output/pidnet_small_cityscapes

For PIDNet checkpoint file is checkpoint_480.pth.tar

For PIDNet + DCN file is DCNV1_600epoch/best.pt

For PIDNet + MDCN file is DCNV2_600epoch/best.pt

!! In the Evaluation phase, change this line in evaluate_model.py

def main(pt_model=False):

If you want to evaluate PIDNet, setting pt_model=False

If you want to evaluate PIDNet + DCN, setting pt_model=True and change checkpoint_path='/home/tanpv/workspace/SelfDC/PIDNet/output/pidnet_small_cityscapes/DCNV1_600epoch/best.pt'

If you want to evaluate PIDNet + MDCN, setting pt_model=True and change checkpoint_path='/home/tanpv/workspace/SelfDC/PIDNet/output/pidnet_small_cityscapes/DCNV2_600epoch/best.pt'

tanphan07 / PIDNet_Deformable_CityScapes