From b2c7686154c6d28f9780c3cec0f7bb86a93e45f6 Mon Sep 17 00:00:00 2001
From: Maciej Wielgosz <maciej.wielgosz@nibio.no>
Date: Tue, 31 Jan 2023 14:52:00 +0100
Subject: [PATCH] update to run on hpc machine
---
.gitignore | 3 ++-
config/partseg.yaml | 2 +-
train_partseg.py | 12 ++++++++++--
3 files changed, 13 insertions(+), 4 deletions(-)
diff --git a/.gitignore b/.gitignore
index 6c1eb22..eb9d135 100644
--- a/.gitignore
+++ b/.gitignore
@@ -2,4 +2,5 @@
__pycache__/
modelnet40_normal_resampled/
outputs/
-log/
\ No newline at end of file
+log/
+data/
\ No newline at end of file
diff --git a/config/partseg.yaml b/config/partseg.yaml
index dccaca0..447566f 100644
--- a/config/partseg.yaml
+++ b/config/partseg.yaml
@@ -1,7 +1,7 @@
batch_size: 16
epoch: 200
learning_rate: 1e-3
-gpu: 1
+gpu: 0
num_point: 1024
optimizer: Adam
weight_decay: 1e-4
diff --git a/train_partseg.py b/train_partseg.py
index a1e9eac..6004991 100644
--- a/train_partseg.py
+++ b/train_partseg.py
@@ -48,9 +48,14 @@ def main(args):
'''HYPER PARAMETER'''
os.environ["CUDA_VISIBLE_DEVICES"] = str(args.gpu)
+
+ # print('GPU available: {}'.format(torch.cuda.is_available()))
logger = logging.getLogger(__name__)
- print(args.pretty())
+ # print(args.pretty())
+
+ # use pretty print to print the config
+
root = hydra.utils.to_absolute_path('data/shapenetcore_partanno_segmentation_benchmark_v0_normal/')
@@ -66,6 +71,9 @@ def main(args):
num_part = args.num_class
shutil.copy(hydra.utils.to_absolute_path('models/{}/model.py'.format(args.model.name)), '.')
+ # print if gpu is available
+ logger.info('GPU available: {}'.format(torch.cuda.is_available()))
+
classifier = getattr(importlib.import_module('models.{}.model'.format(args.model.name)), 'PointTransformerSeg')(args).cuda()
criterion = torch.nn.CrossEntropyLoss()
@@ -202,7 +210,7 @@ def main(args):
mean_shape_ious = np.mean(list(shape_ious.values()))
test_metrics['accuracy'] = total_correct / float(total_seen)
test_metrics['class_avg_accuracy'] = np.mean(
- np.array(total_correct_class) / np.array(total_seen_class, dtype=np.float))
+ np.array(total_correct_class) / np.array(total_seen_class, dtype=np.float32))
for cat in sorted(shape_ious.keys()):
logger.info('eval mIoU of %s %f' % (cat + ' ' * (14 - len(cat)), shape_ious[cat]))
test_metrics['class_avg_iou'] = mean_shape_ious
--
GitLab