From af1fccaa154a4455ea6b5d6c5b1da1e1e427edba Mon Sep 17 00:00:00 2001
From: Gustaf Rydholm <gustaf.rydholm@gmail.com>
Date: Thu, 30 Sep 2021 23:07:54 +0200
Subject: Replace table with image and caption for w&b callback

---
 training/callbacks/wandb_callbacks.py | 40 +++++++++++++++++++++++++++++------
 1 file changed, 33 insertions(+), 7 deletions(-)

(limited to 'training/callbacks')

diff --git a/training/callbacks/wandb_callbacks.py b/training/callbacks/wandb_callbacks.py
index 68e4135..b23f720 100644
--- a/training/callbacks/wandb_callbacks.py
+++ b/training/callbacks/wandb_callbacks.py
@@ -5,6 +5,7 @@ import wandb
 from pytorch_lightning import Callback, LightningModule, Trainer
 from pytorch_lightning.loggers import LoggerCollection, WandbLogger
 from pytorch_lightning.utilities import rank_zero_only
+from torch import nn
 from torch.utils.data import DataLoader
 from torchvision.utils import make_grid
 
@@ -83,8 +84,9 @@ class UploadCheckpointsAsArtifact(Callback):
 class LogTextPredictions(Callback):
     """Logs a validation batch with image to text transcription."""
 
-    def __init__(self, num_samples: int = 8) -> None:
+    def __init__(self, num_samples: int = 8, log_train: bool = False) -> None:
         self.num_samples = num_samples
+        self.log_train = log_train
         self.ready = False
 
     def _log_predictions(
@@ -109,9 +111,8 @@ class LogTextPredictions(Callback):
         logits = pl_module(imgs)
 
         mapping = pl_module.mapping
-        columns = ["image", "prediction", "truth"]
         data = [
-            [wandb.Image(img), mapping.get_text(pred), mapping.get_text(label)]
+            wandb.Image(img, caption=mapping.get_text(pred))
             for img, pred, label in zip(
                 imgs[: self.num_samples],
                 logits[: self.num_samples],
@@ -119,9 +120,7 @@ class LogTextPredictions(Callback):
             )
         ]
 
-        experiment.log(
-            {f"HTR/{experiment.name}/{stage}": wandb.Table(data=data, columns=columns)}
-        )
+        experiment.log({f"HTR/{experiment.name}/{stage}": data})
 
     def on_sanity_check_start(
         self, trainer: Trainer, pl_module: LightningModule
@@ -133,6 +132,15 @@ class LogTextPredictions(Callback):
         """Start executing this callback only after all validation sanity checks end."""
         self.ready = True
 
+    def on_train_epoch_end(self, trainer: Trainer, pl_module: LightningModule) -> None:
+        """Logs predictions on train epoch end."""
+        if not self.log_train:
+            return
+        dataloader = trainer.datamodule.train_dataloader()
+        self._log_predictions(
+            stage="train", trainer=trainer, pl_module=pl_module, dataloader=dataloader
+        )
+
     def on_validation_epoch_end(
         self, trainer: Trainer, pl_module: LightningModule
     ) -> None:
@@ -153,9 +161,13 @@ class LogTextPredictions(Callback):
 class LogReconstuctedImages(Callback):
     """Log reconstructions of images."""
 
-    def __init__(self, num_samples: int = 8) -> None:
+    def __init__(
+        self, num_samples: int = 8, log_train: bool = False, use_sigmoid: bool = False
+    ) -> None:
         self.num_samples = num_samples
+        self.log_train = log_train
         self.ready = False
+        self.sigmoid = nn.Sigmoid() if use_sigmoid else None
 
     def _log_reconstruction(
         self,
@@ -177,6 +189,11 @@ class LogReconstuctedImages(Callback):
 
         imgs = imgs.to(device=pl_module.device)
         reconstructions = pl_module(imgs)[0]
+        reconstructions = (
+            self.sigmoid(reconstructions)
+            if self.sigmoid is not None
+            else reconstructions
+        )
 
         data = [
             wandb.Image(
@@ -200,6 +217,15 @@ class LogReconstuctedImages(Callback):
         """Start executing this callback only after all validation sanity checks end."""
         self.ready = True
 
+    def on_train_epoch_end(self, trainer: Trainer, pl_module: LightningModule) -> None:
+        """Logs predictions on train epoch end."""
+        if not self.log_train:
+            return
+        dataloader = trainer.datamodule.train_dataloader()
+        self._log_reconstruction(
+            stage="train", trainer=trainer, pl_module=pl_module, dataloader=dataloader
+        )
+
     def on_validation_epoch_end(
         self, trainer: Trainer, pl_module: LightningModule
     ) -> None:
-- 
cgit v1.2.3-70-g09d2