feat: reduce the number of parameters in the net

Former-commit-id: 862569b6d284ec8235586b161d8c7055c006f5d8 [formerly f2e672d780df12a398e851f375a238c2d394a3cd]
Former-commit-id: 740b1129a627c488537bb0d0dc7ff73b66fde813

.gitignore

@@ -1,6 +1,10 @@
 .venv/
 .mypy_cache/
 __pycache__/
+
 wandb/
+images/
 
 *.pth
+*.png
+*.jpg

.vscode/launch.json

@@ -12,9 +12,9 @@
       "console": "integratedTerminal",
       "args": [
         "--input",
-        "SM.png",
+        "images/SM.png",
         "--output",
-        "test.png",
+        "output.png",
       ],
       "justMyCode": true
     }

SM.png.REMOVED.git-id

@@ -1 +0,0 @@
-c6d08aa612451072cfe32a3ee086d08342ed9dd9

src/train.py

@@ -18,7 +18,7 @@ from utils.paste import RandomPaste
 
 CHECKPOINT_DIR = Path("./checkpoints/")
 DIR_TRAIN_IMG = Path("/home/lilian/data_disk/lfainsin/val2017")
-DIR_VALID_IMG = Path("/home/lilian/data_disk/lfainsin/smolval2017/")
+DIR_VALID_IMG = Path("/home/lilian/data_disk/lfainsin/smoltrain2017/")
 DIR_SPHERE_IMG = Path("/home/lilian/data_disk/lfainsin/spheres/Images/")
 DIR_SPHERE_MASK = Path("/home/lilian/data_disk/lfainsin/spheres/Masks/")
 
@@ -41,7 +41,7 @@ def get_args():
         dest="batch_size",
         metavar="B",
         type=int,
-        default=16,
+        default=70,
         help="Batch size",
     )
     parser.add_argument(
@@ -92,11 +92,14 @@ def main():
     torch.backends.cudnn.benchmark = True
 
     # 0. Create network
-    net = UNet(n_channels=3, n_classes=args.classes)
+    features = [16, 32, 64, 128]
+    net = UNet(n_channels=args.n_channels, n_classes=args.classes, features=features)
     logging.info(
         f"""Network:
         input channels:  {net.n_channels}
         output channels: {net.n_classes}
+        nb parameters:   {sum(p.numel() for p in net.parameters() if p.requires_grad)}
+        features:        {features}
         """
     )
 
@@ -138,7 +141,7 @@ def main():
     ds_valid = SphereDataset(image_dir=DIR_VALID_IMG, transform=tf_valid)
 
     # 3. Create data loaders
-    loader_args = dict(batch_size=args.batch_size, num_workers=6, pin_memory=True)
+    loader_args = dict(batch_size=args.batch_size, num_workers=8, pin_memory=True)
     train_loader = DataLoader(ds_train, shuffle=True, **loader_args)
     val_loader = DataLoader(ds_valid, shuffle=False, drop_last=True, **loader_args)
 
@@ -159,9 +162,9 @@ def main():
         ),
     )
     wandb.watch(net, log_freq=100)
-    # artifact = wandb.Artifact("model", type="model")
+    artifact_model = wandb.Artifact("model", type="model")
-    # artifact.add_file("model.pth")
+    artifact_model.add_file("model.pth")
-    # run.log_artifact(artifact)
+    run.log_artifact(artifact_model)
 
     logging.info(
         f"""Starting training:
@@ -228,8 +231,7 @@ def main():
             print(f"Train Loss: {train_loss:.3f}, Valid Score: {val_score:3f}")
 
             # save weights when epoch end
-            # torch.save(net.state_dict(), "model.pth")
+            torch.save(net.state_dict(), "model.pth")
-            # run.log_artifact(artifact)
             logging.info(f"model saved!")
 
         run.finish()