| CHECKPOINT=$1 |
| GPU=$2 |
| NAME=$3 |
|
|
| torchrun --nproc_per_node $GPU -m training.main --batch-size=1 \ |
| --model EVA02-CLIP-B-16 --pretrained eva --test-type ade_panoptic --train-data="" \ |
| --val-data data/ADEChallengeData2016/ade20k_panoptic_val.json \ |
| --embed-path metadata/ade_panoptic_clip_hand_craft_EVACLIP_ViTB16.npy \ |
| --val-image-root data/ADEChallengeData2016/images/validation \ |
| --val-segm-root data/ADEChallengeData2016/ade20k_panoptic_val \ |
| --cache-dir $CHECKPOINT --extract-type="v2" \ |
| --name $NAME --downsample-factor 16 --det-image-size 512 |
