feat: use LinearWarmupCosineAnnealingLR

Former-commit-id: a7292fe2b0898513fd0e913c2ae352a187f05b12 [formerly 432664f5bd6c3f8f54c221e4d7cc8853d08ea55b]
Former-commit-id: b777bd5053a005ec9b4da7db19a4dbe2a1ba41fd

src/mrcnn/module.py

@@ -4,6 +4,7 @@ import pytorch_lightning as pl
 import torch
 import torchvision
 import wandb
+from pl_bolts.optimizers.lr_scheduler import LinearWarmupCosineAnnealingLR
 from torchmetrics.detection.mean_ap import MeanAveragePrecision
 from torchvision.models.detection.faster_rcnn import FastRCNNPredictor
 from torchvision.models.detection.mask_rcnn import (
@@ -43,8 +44,8 @@ class MRCNNModule(pl.LightningModule):
         # Network
         self.model = get_model_instance_segmentation(n_classes)
 
-        # onnx
-        self.example_input_array = torch.randn(1, 3, 1024, 1024, requires_grad=True)
+        # onnx export
+        self.example_input_array = torch.randn(1, 3, 1024, 1024, requires_grad=True).half()
 
     def forward(self, imgs):
         self.model.eval()
@@ -78,22 +79,9 @@ class MRCNNModule(pl.LightningModule):
             target["masks"] = target["masks"].squeeze(1).bool()
         self.metric.update(preds, targets)
 
-        # compute validation loss
-        self.model.train()
-        loss_dict = self.model(images, targets)
-        loss_dict = {f"valid/{key}": val for key, val in loss_dict.items()}
-        loss_dict["valid/loss"] = sum(loss_dict.values())
-        self.model.eval()
-
-        return loss_dict
+        return preds
 
     def validation_epoch_end(self, outputs):
-        # log validation loss
-        loss_dict = {
-            k: torch.stack([d[k] for d in outputs]).mean() for k in outputs[0].keys()
-        }  # TODO: update un dict object
-        self.log_dict(loss_dict)
-
         # log metrics
         metric_dict = self.metric.compute()
         metric_dict = {f"valid/{key}": val for key, val in metric_dict.items()}
@@ -103,24 +91,20 @@ class MRCNNModule(pl.LightningModule):
         optimizer = torch.optim.Adam(
             self.parameters(),
             lr=wandb.config.LEARNING_RATE,
-            # momentum=wandb.config.MOMENTUM,
-            # weight_decay=wandb.config.WEIGHT_DECAY,
+            momentum=wandb.config.MOMENTUM,
+            weight_decay=wandb.config.WEIGHT_DECAY,
         )
 
-        # scheduler = torch.optim.lr_scheduler.CosineAnnealingWarmRestarts(
-        #     optimizer,
-        #     T_0=3,
-        #     T_mult=1,
-        #     lr=wandb.config.LEARNING_RATE_MIN,
-        #     verbose=True,
-        # )
+        scheduler = LinearWarmupCosineAnnealingLR(
+            optimizer,
+            warmup_epochs=10,
+            max_epochs=40,
+        )
 
-        # return {
-        #     "optimizer": optimizer,
-        #     "lr_scheduler": {
-        #         "scheduler": scheduler,
-        #         "monitor": "val_accuracy",
-        #     },
-        # }
-
-        return optimizer
+        return {
+            "optimizer": optimizer,
+            "lr_scheduler": {
+                "scheduler": scheduler,
+                "monitor": "map",
+            },
+        }