export CUDA=0 export BASE_TEST_DATA_DIR="datasets/eval/" export CHECKPOINT_DIR="jingheya/lotus-depth-g-v2-1-disparity" export OUTPUT_DIR="output/Depth_G_Eval" export TASK_NAME="depth" export MODE="generation" 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 # The defualt `processing_res` is set in the configuration file of each dataset.