pstage_01_image_classification

Getting Started

Dependencies

• torch==1.6.0
• torchvision==0.7.0

Install Requirements

• pip install -r requirements.txt

Training

• python train.py --augmentation CustomAugmentation --resize 256 192 --model densenet --optimizer AdamW --lr 0.00009 --criterion focal --lr_decay_step 1 --pretrained True --early_stop 1 --LR_scheduler StepLR --steplr_gamma 0.1
• "transform": "Compose(\n
  • Resize(size=[256, 192],
  • interpolation=PIL.Image.BILINEAR)\n
  • RandomHorizontalFlip(p=0.5)\n
  • ToTensor()\n
  • Normalize(mean=(0.548, 0.504, 0.479), std=(0.237, 0.247, 0.246))\n)"

Inference

• python inference.py --resize 256 192 --model densenet --model_dir ./model/exp_number

Evaluation

• SM_GROUND_TRUTH_DIR=[GT dir] SM_OUTPUT_DATA_DIR=[inference output dir] python evaluation.py