-
Notifications
You must be signed in to change notification settings - Fork 0
/
eval_cor_modelnet40c.sh
35 lines (23 loc) · 1.48 KB
/
eval_cor_modelnet40c.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
# evaluation on the original ModelNet40-C dataset
# models trained end-to-end
if [ ! -d "modelnetc"]; then
mkdir "modelnetc"
fi
for model in 'pct' 'pctc' 'pctcmean' 'peat' 'peatmean' 'rscnn' 'pointnet2' 'dgcnn' 'pemax' 'pemean' 'pemedian' 'pointnetmean' 'pointnet' 'pointnetmlp3mean'; do
for cor in 'uniform' 'gaussian' 'background' 'impulse' 'upsampling' 'distortion_rbf' 'distortion_rbf_inv' 'density' 'density_inc' 'shear' 'rotation' 'cutout' 'distortion' 'occlusion' 'lidar'; do
for sev in 1 2 3 4 5; do
CUDA_VISIBLE_DEVICES=1 python main.py --entry test --model-path runs/dgcnn_${model}_run_1/model_best_test.pth --exp-config configs/corruption/${model}.yaml --iid 0 --severity ${sev} --corruption ${cor} --output ./modelnetc/${model}_none_${cor}_${sev}.txt
done
done
done
# analytical or random initialized per-point embedding
if [ ! -d "modelnetc" ]; then
mkdir "modelnetc"
fi
for model in 'pctc' 'pctcmean' 'peat' 'peatmean' 'pointnetmean' 'pointnet' 'pointnetmlp3mean'; do #
for cor in 'uniform' 'gaussian' 'background' 'impulse' 'upsampling' 'distortion_rbf' 'distortion_rbf_inv' 'density' 'density_inc' 'shear' 'rotation' 'cutout' 'distortion' 'occlusion' 'lidar'; do
for sev in 1 2 3 4 5; do
CUDA_VISIBLE_DEVICES=0 python main.py --entry test --model-path runs/dgcnn_${model}_run_1_random/model_best_test.pth --exp-config configs/corruption/${model}.yaml --iid 0 --severity ${sev} --corruption ${cor} --output ./modelnetc/${model}+random_none_${cor}_${sev}.txt
done
done
done