Lotus/eval_scripts/eval-normal-d.sh

export CUDA=0

export BASE_TEST_DATA_DIR="datasets/eval/"

export CHECKPOINT_DIR="jingheya/lotus-normal-d-v1-1"
export OUTPUT_DIR="output/Normal_D_Eval"
export TASK_NAME="normal"

export MODE="regression"

CUDA_VISIBLE_DEVICES=$CUDA python eval.py \
        --pretrained_model_name_or_path=$CHECKPOINT_DIR \
        --prediction_type="sample" \
        --seed=42 \
        --half_precision \
        --base_test_data_dir=$BASE_TEST_DATA_DIR \
        --task_name=$TASK_NAME \
        --mode=$MODE \
        --output_dir=$OUTPUT_DIR \
        --disparity
        # You can set `processing_res` for high-resolution images. Default: `--processing_res=None`.