From fd78121725bc15789cbddb86f4e2b87ae50c52ed Mon Sep 17 00:00:00 2001
From: Maciej Wielgosz <maciej.wielgosz@nibio.no>
Date: Fri, 24 Mar 2023 11:20:17 +0100
Subject: [PATCH] update dgcnn loader for to small final batch
---
dgcnn/dataset_run.ipynb | 16 +++++++++++-----
dgcnn/dgcnn_train.py | 20 ++++++++++++--------
dgcnn/find_missing_data.py | 20 ++++++++++++++++++++
3 files changed, 43 insertions(+), 13 deletions(-)
create mode 100644 dgcnn/find_missing_data.py
diff --git a/dgcnn/dataset_run.ipynb b/dgcnn/dataset_run.ipynb
index 5265d6c..23ad7b8 100644
--- a/dgcnn/dataset_run.ipynb
+++ b/dgcnn/dataset_run.ipynb
@@ -39,17 +39,23 @@
},
{
"cell_type": "code",
- "execution_count": 8,
+ "execution_count": 9,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "tensor([[-0.0148, -1.7624, -2.2409, 0.8483, 0.4169],\n",
- " [-0.3802, -0.0168, -1.9958, 0.9351, -1.0900],\n",
- " [ 0.3501, -1.3573, -1.8246, -1.2850, 0.2785]], requires_grad=True)\n",
- "tensor([0, 0, 1])\n"
+ "tensor([[-0.1308, -0.9843, 0.0595, -1.4868, -0.0280],\n",
+ " [ 1.6206, -0.0315, 0.6667, 2.3133, -1.0334],\n",
+ " [ 0.3246, 0.6226, -0.9101, 0.1938, 0.6888]], requires_grad=True)\n",
+ "tensor([3, 4, 0])\n",
+ "tensor([[-1.2085, 0.0789, -0.9984, 0.3746, -0.6782],\n",
+ " [-1.1200, -0.0271, -0.6976, -0.2629, 1.4103],\n",
+ " [-2.2474, 0.5892, 0.8203, -1.2913, -1.5551]], requires_grad=True)\n",
+ "tensor([[0.2879, 0.0775, 0.0657, 0.4815, 0.0875],\n",
+ " [0.2927, 0.2297, 0.3383, 0.1158, 0.0235],\n",
+ " [0.2733, 0.3353, 0.1628, 0.0173, 0.2113]])\n"
]
}
],
diff --git a/dgcnn/dgcnn_train.py b/dgcnn/dgcnn_train.py
index 08aeea0..59f502e 100644
--- a/dgcnn/dgcnn_train.py
+++ b/dgcnn/dgcnn_train.py
@@ -21,20 +21,22 @@ def train():
# get data
shapenet_data = ShapenetDataDgcnn(
root='/home/nibio/mutable-outside-world/code/oracle_gpu_runs/data/shapenet',
- npoints=32,
+ npoints=256,
return_cls_label=True,
- small_data=True,
+ small_data=False,
small_data_size=1000,
- split='train'
+ just_one_class=False,
+ split='train',
+ norm=True
)
# create a dataloader
dataloader = torch.utils.data.DataLoader(
shapenet_data,
- batch_size=8,
- shuffle=False,
- num_workers=4,
- drop_last=False
+ batch_size=4,
+ shuffle=True,
+ num_workers=8,
+ drop_last=True
)
# create a optimizer
@@ -43,7 +45,7 @@ def train():
# create a config wandb
wandb.config.update({
"batch_size": 8,
- "learning_rate": 0.0001,
+ "learning_rate": 0.01,
"optimizer": "Adam",
"loss_function": "cross_entropy"
})
@@ -57,6 +59,8 @@ def train():
print(f"Epoch: {epoch}")
wandb.log({"epoch": epoch})
for i, data in enumerate(dataloader, 0):
+ # print(f"Batch: {i}")
+
points, _, class_name = data
points = points.cuda()
class_name = class_name.cuda()
diff --git a/dgcnn/find_missing_data.py b/dgcnn/find_missing_data.py
new file mode 100644
index 0000000..d9bcc64
--- /dev/null
+++ b/dgcnn/find_missing_data.py
@@ -0,0 +1,20 @@
+from tqdm import tqdm
+from shapenet_data_dgcnn import ShapenetDataDgcnn
+
+shapenet_data = ShapenetDataDgcnn(
+ root='/home/nibio/mutable-outside-world/code/oracle_gpu_runs/data/shapenet',
+ npoints=128,
+ return_cls_label=True,
+ small_data=False,
+ small_data_size=1000,
+ just_one_class=False,
+ split='train',
+ norm=True
+ )
+
+# read the data one by one and check if exists
+
+for i in tqdm(range(len(shapenet_data))):
+ data = shapenet_data[i]
+ if data[0].shape[0] != 128:
+ print(f"Data is None: {i}")
\ No newline at end of file
--
GitLab