why the validation result so low?

[miaoxiaoting]

我修改了gkt_nuscenes_vehicle_kernel_7x1_setting1.yaml中BEV h的尺寸，可是训练过程中，val 指标特别低，基本不变 ，不知道为什么 data: bev:

h: 400 # TODO 程序跑不通

h: 200
w: 200
# h_meters: 100.0 # TODO
h_meters: 50.0
w_meters: 50.0

python scripts/train.py +experiment=gkt_nuscenes_vehicle_kernel_7x1_setting1.yaml \

data.dataset_dir=/GKT/data/nuscenes/mini data.labels_dir=/GKT/data/nuscenes/cvt_labels_nuscenes_v2

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓

┃ Validate metric ┃ DataLoader 0 ┃

┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩

│ train/metrics/iou@0.40 │ 0.9998205304145813 │

│ train/metrics/iou@0.50 │ 0.9998857975006104 │

│ train/metrics/iou_with_occlusions@0.40 │ 0.7885284423828125 │

│ train/metrics/iou_with_occlusions@0.50 │ 0.7882291674613953 │

│ val/loss │ 0.17532268166542053 │

│ val/loss/center │ 0.003345666453242302 │

│ val/loss/visible │ 0.17498812079429626 │

│ val/metrics/iou@0.40 │ 0.051858142018318176 │

│ val/metrics/iou@0.50 │ 0.0490204356610775 │

│ val/metrics/iou_with_occlusions@0.40 │ 0.037373993545770645 │

│ val/metrics/iou_with_occlusions@0.50 │ 0.03504466265439987 │

└────────────────────────────────────────┴────────────────────────────────────────┘