Save weights/Generate samples improvements

* "Save weights" and "Generate samples" happen immediately * Generating samples won't try to load every image in the dataset when "[filewords]" is used * Generating samples handles exceptions reading images for "[filewords]"
RossM · Jan 20, 2024 · 05b5ba2 · 05b5ba2
1 parent 39acf55
commit 05b5ba2
Show file tree

Hide file tree

Showing 2 changed files with 35 additions and 25 deletions.
diff --git a/dreambooth/dataset/sample_dataset.py b/dreambooth/dataset/sample_dataset.py
@@ -38,18 +38,27 @@ def __init__(self, config: DreamboothConfig):
             elif "[filewords]" in sample_prompt:
                 prompts = []
                 images = get_images(concept.instance_data_dir)
+                random.shuffle(images)
                 getter = FilenameTextGetter(shuffle_tags)
+                selected = 0
                 for image in images:
-                    file_text = getter.read_text(image)
-                    prompt = getter.create_text(sample_prompt, file_text, concept, False)
-                    img = Image.open(image)
-                    res = img.size
-                    closest = closest_resolution(res[0], res[1], bucket_resos)
-                    prompts.append((prompt, closest))
+                    try:
+                        file_text = getter.read_text(image)
+                        prompt = getter.create_text(sample_prompt, file_text, concept, False)
+                        img = Image.open(image)
+                        res = img.size
+                        closest = closest_resolution(res[0], res[1], bucket_resos)
+                        prompts.append((prompt, closest))
+                        selected += 1
+                        if selected >= required:
+                            break
+                    except:
+                        pass
             else:
                 prompts = [(sample_prompt, (config.resolution, config.resolution))]
+            random.shuffle(prompts)
             for i in range(required):
-                pi = random.choice(prompts)
+                pi = prompts[i % len(prompts)]
                 pd = PromptData(
                     prompt=pi[0],
                     negative_prompt=neg,

diff --git a/dreambooth/train_dreambooth.py b/dreambooth/train_dreambooth.py
@@ -1012,7 +1012,14 @@ def check_save(is_epoch_check=False):
                 save_model = False
                 save_lora = False
 
-                if not save_canceled and not save_completed:
+                if save_canceled or save_completed:
+                    logger.debug("\nSave completed/canceled.")
+                    if global_step > 0:
+                        save_image = True
+                        save_model = True
+                        if args.use_lora:
+                            save_lora = True
+                elif is_epoch_check:
                     # Check to see if the number of epochs since last save is gt the interval
                     if 0 < save_model_interval <= session_epoch - last_model_save:
                         save_model = True
@@ -1025,26 +1032,17 @@ def check_save(is_epoch_check=False):
                         save_image = True
                         last_image_save = session_epoch
 
-                else:
-                    logger.debug("\nSave completed/canceled.")
-                    if global_step > 0:
-                        save_image = True
-                        save_model = True
-                        if args.use_lora:
-                            save_lora = True
-
                 save_snapshot = False
 
-                if is_epoch_check:
-                    if shared.status.do_save_samples:
-                        save_image = True
-                        shared.status.do_save_samples = False
+                if shared.status.do_save_samples:
+                    save_image = True
+                    shared.status.do_save_samples = False
 
-                    if shared.status.do_save_model:
-                        if args.use_lora:
-                            save_lora = True
-                        save_model = True
-                        shared.status.do_save_model = False
+                if shared.status.do_save_model:
+                    if args.use_lora:
+                        save_lora = True
+                    save_model = True
+                    shared.status.do_save_model = False
 
                 save_checkpoint = False
                 if save_model:
@@ -1911,6 +1909,9 @@ def lora_save_function(weights, filename):
                             status_handler.end(status.textinfo)
                         break
 
+                    if status.do_save_model or status.do_save_samples:
+                        check_save(False)
+
                 accelerator.wait_for_everyone()
 
                 args.epoch += 1