Merge branch 'master' into fix/train-preprocess-keep-ratio

modules/textual_inversion/dataset.py

@@ -83,7 +83,7 @@ class PersonalizedBase(Dataset):
 
             self.dataset.append(entry)
 
-        assert len(self.dataset) > 1, "No images have been found in the dataset."
+        assert len(self.dataset) > 0, "No images have been found in the dataset."
         self.length = len(self.dataset) * repeats // batch_size
 
         self.initial_indexes = np.arange(len(self.dataset))
@@ -91,7 +91,7 @@ class PersonalizedBase(Dataset):
         self.shuffle()
 
     def shuffle(self):
-        self.indexes = self.initial_indexes[torch.randperm(self.initial_indexes.shape[0])]
+        self.indexes = self.initial_indexes[torch.randperm(self.initial_indexes.shape[0]).numpy()]
 
     def create_text(self, filename_text):
         text = random.choice(self.lines)

modules/textual_inversion/image_embedding.py

@@ -5,6 +5,7 @@ import zlib
 from PIL import Image, PngImagePlugin, ImageDraw, ImageFont
 from fonts.ttf import Roboto
 import torch
+from modules.shared import opts
 
 
 class EmbeddingEncoder(json.JSONEncoder):
@@ -133,7 +134,7 @@ def caption_image_overlay(srcimage, title, footerLeft, footerMid, footerRight, t
     from math import cos
 
     image = srcimage.copy()
-
+    fontsize = 32
     if textfont is None:
         try:
             textfont = ImageFont.truetype(opts.font or Roboto, fontsize)
@@ -150,7 +151,7 @@ def caption_image_overlay(srcimage, title, footerLeft, footerMid, footerRight, t
     image = Image.alpha_composite(image.convert('RGBA'), gradient.resize(image.size))
 
     draw = ImageDraw.Draw(image)
-    fontsize = 32
+
     font = ImageFont.truetype(textfont, fontsize)
     padding = 10
 

modules/textual_inversion/preprocess.py

@@ -12,7 +12,7 @@ if cmd_opts.deepdanbooru:
     import modules.deepbooru as deepbooru
 
 
-def preprocess(process_src, process_dst, process_width, process_height, process_flip, process_split, process_caption, process_caption_deepbooru=False, split_threshold=0.5, overlap_ratio=0.2):
+def preprocess(process_src, process_dst, process_width, process_height, preprocess_txt_action, process_flip, process_split, process_caption, process_caption_deepbooru=False, split_threshold=0.5, overlap_ratio=0.2):
     try:
         if process_caption:
             shared.interrogator.load()
@@ -22,7 +22,7 @@ def preprocess(process_src, process_dst, process_width, process_height, process_
             db_opts[deepbooru.OPT_INCLUDE_RANKS] = False
             deepbooru.create_deepbooru_process(opts.interrogate_deepbooru_score_threshold, db_opts)
 
-        preprocess_work(process_src, process_dst, process_width, process_height, process_flip, process_split, process_caption, process_caption_deepbooru, split_threshold, overlap_ratio)
+        preprocess_work(process_src, process_dst, process_width, process_height, preprocess_txt_action, process_flip, process_split, process_caption, process_caption_deepbooru, split_threshold, overlap_ratio)
 
     finally:
 
@@ -34,7 +34,7 @@ def preprocess(process_src, process_dst, process_width, process_height, process_
 
 
 
-def preprocess_work(process_src, process_dst, process_width, process_height, process_flip, process_split, process_caption, process_caption_deepbooru=False, split_threshold=0.5, overlap_ratio=0.2):
+def preprocess_work(process_src, process_dst, process_width, process_height, preprocess_txt_action, process_flip, process_split, process_caption, process_caption_deepbooru=False, split_threshold=0.5, overlap_ratio=0.2):
     width = process_width
     height = process_height
     src = os.path.abspath(process_src)
@@ -51,7 +51,7 @@ def preprocess_work(process_src, process_dst, process_width, process_height, pro
     shared.state.textinfo = "Preprocessing..."
     shared.state.job_count = len(files)
 
-    def save_pic_with_caption(image, index):
+    def save_pic_with_caption(image, index, existing_caption=None):
         caption = ""
 
         if process_caption:
@@ -69,17 +69,26 @@ def preprocess_work(process_src, process_dst, process_width, process_height, pro
         basename = f"{index:05}-{subindex[0]}-{filename_part}"
         image.save(os.path.join(dst, f"{basename}.png"))
 
+        if preprocess_txt_action == 'prepend' and existing_caption:
+            caption = existing_caption + ' ' + caption
+        elif preprocess_txt_action == 'append' and existing_caption:
+            caption = caption + ' ' + existing_caption
+        elif preprocess_txt_action == 'copy' and existing_caption:
+            caption = existing_caption
+
+        caption = caption.strip()
+        
         if len(caption) > 0:
             with open(os.path.join(dst, f"{basename}.txt"), "w", encoding="utf8") as file:
                 file.write(caption)
 
         subindex[0] += 1
 
-    def save_pic(image, index):
-        save_pic_with_caption(image, index)
+    def save_pic(image, index, existing_caption=None):
+        save_pic_with_caption(image, index, existing_caption=existing_caption)
 
         if process_flip:
-            save_pic_with_caption(ImageOps.mirror(image), index)
+            save_pic_with_caption(ImageOps.mirror(image), index, existing_caption=existing_caption)
 
     def split_pic(image, inverse_xy):
         if inverse_xy:
@@ -112,6 +121,13 @@ def preprocess_work(process_src, process_dst, process_width, process_height, pro
         except Exception:
             continue
 
+        existing_caption = None
+
+        try:
+            existing_caption = open(os.path.splitext(filename)[0] + '.txt', 'r').read()
+        except Exception as e:
+            print(e)
+
         if shared.state.interrupted:
             break
 
@@ -124,9 +140,9 @@ def preprocess_work(process_src, process_dst, process_width, process_height, pro
 
         if process_split and ratio < 1.0 and ratio <= split_threshold:
             for splitted in split_pic(img, inverse_xy):
-                save_pic(splitted, index)
+                save_pic(splitted, index, existing_caption=existing_caption)
         else:
             img = images.resize_image(1, img, width, height)
-            save_pic(img, index)
+            save_pic(img, index, existing_caption=existing_caption)
 
         shared.state.nextjob()

modules/textual_inversion/textual_inversion.py

@@ -153,7 +153,7 @@ class EmbeddingDatabase:
         return None, None
 
 
-def create_embedding(name, num_vectors_per_token, init_text='*'):
+def create_embedding(name, num_vectors_per_token, overwrite_old, init_text='*'):
     cond_model = shared.sd_model.cond_stage_model
     embedding_layer = cond_model.wrapped.transformer.text_model.embeddings
 
@@ -165,7 +165,8 @@ def create_embedding(name, num_vectors_per_token, init_text='*'):
         vec[i] = embedded[i * int(embedded.shape[0]) // num_vectors_per_token]
 
     fn = os.path.join(shared.cmd_opts.embeddings_dir, f"{name}.pt")
-    assert not os.path.exists(fn), f"file {fn} already exists"
+    if not overwrite_old:
+        assert not os.path.exists(fn), f"file {fn} already exists"
 
     embedding = Embedding(vec, name)
     embedding.step = 0
@@ -275,6 +276,7 @@ def train_embedding(embedding_name, learn_rate, batch_size, data_root, log_direc
             loss.backward()
             optimizer.step()
 
+
         epoch_num = embedding.step // len(ds)
         epoch_step = embedding.step - (epoch_num * len(ds)) + 1
 

modules/textual_inversion/ui.py

@@ -7,8 +7,8 @@ import modules.textual_inversion.preprocess
 from modules import sd_hijack, shared
 
 
-def create_embedding(name, initialization_text, nvpt):
-    filename = modules.textual_inversion.textual_inversion.create_embedding(name, nvpt, init_text=initialization_text)
+def create_embedding(name, initialization_text, nvpt, overwrite_old):
+    filename = modules.textual_inversion.textual_inversion.create_embedding(name, nvpt, overwrite_old, init_text=initialization_text)
 
     sd_hijack.model_hijack.embedding_db.load_textual_inversion_embeddings()