evaluation metrics was very bad

[dong-hub-png]

Hello, thank you for your contribution.I trained bevfusion-mit using my own dataset, and after quantization, an anomaly occurred where all evaluation values ​​were 0.
`mAP: 0.0000
mATE: 1.0448
mASE: 0.9949
mAOE: 1.0000
NDS: 0.0006
Eval time: 1.2s

Per-class results:
Object Class AP ATE ASE AOE
Car 0.000 1.000 1.000 1.000
igv_without_box 0.000 1.000 1.000 1.000
igv_with_box 0.000 1.000 1.000 1.000
pillar 0.000 1.436 0.979 nan
cones 0.000 1.012 0.970 nan
pedestrian 0.000 1.000 1.000 nan
truck_head 0.000 1.000 1.000 1.000
truck_tail_without_box 0.000 1.000 1.000 1.000
truck_tail_with_halfbox 0.000 1.000 1.000 1.000
truck_tail_with_box 0.000 1.000 1.000 1.000 Where might this problem occur?The test results of the model I trained are:[>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 275/275, 2.8 task/s, elapsed: 99s, ETA: 0s
mAP: 0.8512
mATE: 0.1395
mASE: 0.1361
mAOE: 0.1294
NDS: 0.8564
Eval time: 2.1s

Per-class results:
Object Class AP ATE ASE AOE
Car 0.993 0.076 0.074 0.028
igv_without_box 0.326 0.092 0.074 0.013
igv_with_box 0.978 0.143 0.070 0.045
pillar 0.773 0.040 0.174 nan
cones 0.640 0.283 0.352 nan
pedestrian 0.995 0.076 0.217 nan
truck_head 0.971 0.166 0.124 0.263
truck_tail_without_box 0.939 0.173 0.115 0.159
truck_tail_with_halfbox 0.920 0.186 0.075 0.124
truck_tail_with_box 0.978 0.160 0.085 0.274 `