add joytag

Signed-off-by: Vladimir Mandic <mandic00@live.com>

modules/interrogate/deepbooru.py

@@ -81,8 +81,8 @@ class DeepDanbooru:
                 tag_outformat = tag_outformat.replace('_', ' ')
             if shared.opts.deepbooru_escape:
                 tag_outformat = re.sub(re_special, r'\\\1', tag_outformat)
-            if shared.opts.deepbooru_clip_score and not force_disable_ranks:
-                tag_outformat = f"({tag_outformat}:{probability:.3f})"
+            if shared.opts.interrogate_score and not force_disable_ranks:
+                tag_outformat = f"({tag_outformat}:{probability:.2f})"
             res.append(tag_outformat)
         if len(res) > shared.opts.deepbooru_max_tags:
             res = res[:shared.opts.deepbooru_max_tags]

modules/interrogate/openclip.py

@@ -213,8 +213,8 @@ class InterrogateModels:
                 for _name, topn, items in self.categories():
                     matches = self.rank(image_features, items, top_count=topn)
                     for match, score in matches:
-                        if shared.opts.interrogate_clip_score:
-                            res += f", ({match}:{score/100:.3f})"
+                        if shared.opts.interrogate_score:
+                            res += f", ({match}:{score/100:.2f})"
                         else:
                             res += f", {match}"
         except Exception as e:

modules/interrogate/vqa.py

@@ -37,6 +37,7 @@ vlm_models = {
     "Microsoft GIT VQA Large": "microsoft/git-large-vqav2", # 1.6GB
     "ToriiGate 0.4 2B": "Minthy/ToriiGate-v0.4-2B",
     "ViLT Base": "dandelin/vilt-b32-finetuned-vqa", # 0.5GB
+    "JoyTag": "fancyfeast/joytag",
 }
 vlm_prompts = [
     '<CAPTION>',
@@ -344,6 +345,9 @@ def interrogate(question, image, model_name):
             answer = qwen(question, image, vqa_model)
         elif 'smol' in vqa_model.lower():
             answer = smol(question, image, vqa_model)
+        elif 'joytag' in vqa_model.lower():
+            from modules.interrogate import joytag
+            answer = joytag.predict(image)
         else:
             answer = 'unknown model'
     except Exception as e:

modules/shared.py

@@ -910,12 +910,12 @@ options_templates.update(options_section(('control', "Control Options"), {
 options_templates.update(options_section(('interrogate', "Interrogate"), {
     "interrogate_default_type": OptionInfo("OpenCLiP", "Default type", gr.Radio, {"choices": ["OpenCLiP", "VLM", "DeepBooru"]}),
     "interrogate_offload": OptionInfo(True, "Interrogate: offload models "),
+    "interrogate_score": OptionInfo(False, "Include scores in results when available"),
 
     "interrogate_clip_sep": OptionInfo("<h2>OpenCLiP</h2>", "", gr.HTML),
     "interrogate_clip_model": OptionInfo("ViT-L-14/openai", "CLiP: default model", gr.Dropdown, lambda: {"choices": get_clip_models()}, refresh=refresh_clip_models),
     "interrogate_clip_mode": OptionInfo(caption_types[0], "CLiP: default mode", gr.Dropdown, {"choices": caption_types}),
     "interrogate_blip_model": OptionInfo(list(caption_models)[0], "CLiP: default captioner", gr.Dropdown, {"choices": list(caption_models)}),
-    "interrogate_clip_score": OptionInfo(False, "CLiP: include scores in results"),
     "interrogate_clip_num_beams": OptionInfo(1, "CLiP: num beams", gr.Slider, {"minimum": 1, "maximum": 16, "step": 1}),
     "interrogate_clip_min_length": OptionInfo(32, "CLiP: min length", gr.Slider, {"minimum": 1, "maximum": 128, "step": 1}),
     "interrogate_clip_max_length": OptionInfo(74, "CLiP: max length", gr.Slider, {"minimum": 1, "maximum": 512, "step": 1}),