Assign unused images in validation dataset to train dataset

4 files changed, 26 insertions, 9 deletions

diff --git a/data/dreambooth/csv.py b/data/dreambooth/csv.py
index 1676d35..4087226 100644
--- a/data/dreambooth/csv.py
+++ b/data/dreambooth/csv.py
@@ -21,6 +21,7 @@ class CSVDataModule(pl.LightningDataModule):
                  interpolation="bicubic",
                  identifier="*",
                  center_crop=False,
+                 valid_set_size=None,
                  collate_fn=None):
         super().__init__()
 
@@ -39,6 +40,7 @@ class CSVDataModule(pl.LightningDataModule):
         self.identifier = identifier
         self.center_crop = center_crop
         self.interpolation = interpolation
+        self.valid_set_size = valid_set_size
         self.collate_fn = collate_fn
         self.batch_size = batch_size
 
@@ -50,8 +52,11 @@ class CSVDataModule(pl.LightningDataModule):
         self.data_full = [(img, cap) for img, cap, skip in zip(image_paths, captions, skips) if skip != "x"]
 
     def setup(self, stage=None):
-        train_set_size = int(len(self.data_full) * 0.8)
-        valid_set_size = len(self.data_full) - train_set_size
+        valid_set_size = int(len(self.data_full) * 0.2)
+        if self.valid_set_size:
+            valid_set_size = math.min(valid_set_size, self.valid_set_size)
+        train_set_size = len(self.data_full) - valid_set_size
+
         self.data_train, self.data_val = random_split(self.data_full, [train_set_size, valid_set_size])
 
         train_dataset = CSVDataset(self.data_train, self.tokenizer, instance_prompt=self.instance_prompt,
diff --git a/data/textual_inversion/csv.py b/data/textual_inversion/csv.py
index f306c7a..e082511 100644
--- a/data/textual_inversion/csv.py
+++ b/data/textual_inversion/csv.py
@@ -18,7 +18,8 @@ class CSVDataModule(pl.LightningDataModule):
                  repeats=100,
                  interpolation="bicubic",
                  placeholder_token="*",
-                 center_crop=False):
+                 center_crop=False,
+                 valid_set_size=None):
         super().__init__()
 
         self.data_file = Path(data_file)
@@ -33,6 +34,7 @@ class CSVDataModule(pl.LightningDataModule):
         self.placeholder_token = placeholder_token
         self.center_crop = center_crop
         self.interpolation = interpolation
+        self.valid_set_size = valid_set_size
 
         self.batch_size = batch_size
 
@@ -44,8 +46,11 @@ class CSVDataModule(pl.LightningDataModule):
         self.data_full = [(img, cap) for img, cap, skip in zip(image_paths, captions, skips) if skip != "x"]
 
     def setup(self, stage=None):
-        train_set_size = int(len(self.data_full) * 0.8)
-        valid_set_size = len(self.data_full) - train_set_size
+        valid_set_size = int(len(self.data_full) * 0.2)
+        if self.valid_set_size:
+            valid_set_size = math.min(valid_set_size, self.valid_set_size)
+        train_set_size = len(self.data_full) - valid_set_size
+
         self.data_train, self.data_val = random_split(self.data_full, [train_set_size, valid_set_size])
 
         train_dataset = CSVDataset(self.data_train, self.tokenizer, size=self.size, repeats=self.repeats, interpolation=self.interpolation,
diff --git a/dreambooth.py b/dreambooth.py
index 744d1bc..88cd0da 100644
--- a/dreambooth.py
+++ b/dreambooth.py
@@ -638,6 +638,7 @@ def main():
         identifier=args.identifier,
         repeats=args.repeats,
         center_crop=args.center_crop,
+        valid_set_size=args.sample_batch_size*args.stable_sample_batches,
         collate_fn=collate_fn)
 
     datamodule.prepare_data()
@@ -658,7 +659,7 @@ def main():
         sample_batch_size=args.sample_batch_size,
         random_sample_batches=args.random_sample_batches,
         stable_sample_batches=args.stable_sample_batches,
-        seed=args.seed
+        seed=args.seed or torch.random.seed()
     )
 
     # Scheduler and math around the number of training steps.
diff --git a/textual_inversion.py b/textual_inversion.py
index 7a7d7fc..fa6214e 100644
--- a/textual_inversion.py
+++ b/textual_inversion.py
@@ -604,9 +604,15 @@ def main():
     )
 
     datamodule = CSVDataModule(
-        data_file=args.train_data_file, batch_size=args.train_batch_size, tokenizer=tokenizer,
-        size=args.resolution, placeholder_token=args.placeholder_token, repeats=args.repeats,
-        center_crop=args.center_crop)
+        data_file=args.train_data_file,
+        batch_size=args.train_batch_size,
+        tokenizer=tokenizer,
+        size=args.resolution,
+        placeholder_token=args.placeholder_token,
+        repeats=args.repeats,
+        center_crop=args.center_crop,
+        valid_set_size=args.sample_batch_size*args.stable_sample_batches
+    )
 
     datamodule.prepare_data()
     datamodule.setup()