stable-diffusion-webui/modules/interrogate.py

import os
import sys
from collections import namedtuple
from pathlib import Path
import re

import torch
import torch.hub

from torchvision import transforms
from torchvision.transforms.functional import InterpolationMode

from modules import devices, paths, shared, lowvram, modelloader, errors, torch_utils

blip_image_eval_size = 384
clip_model_name = 'ViT-L/14'

Category = namedtuple("Category", ["name", "topn", "items"])

re_topn = re.compile(r"\.top(\d+)$")

def category_types():
    return [f.stem for f in Path(shared.interrogator.content_dir).glob('*.txt')]


def download_default_clip_interrogate_categories(content_dir):
    print("Downloading CLIP categories...")

    tmpdir = f"{content_dir}_tmp"
    category_types = ["artists", "flavors", "mediums", "movements"]

    try:
        os.makedirs(tmpdir, exist_ok=True)
        for category_type in category_types:
            torch.hub.download_url_to_file(f"https://raw.githubusercontent.com/pharmapsychotic/clip-interrogator/main/clip_interrogator/data/{category_type}.txt", os.path.join(tmpdir, f"{category_type}.txt"))
        os.rename(tmpdir, content_dir)

    except Exception as e:
        errors.display(e, "downloading default CLIP interrogate categories")
    finally:
        if os.path.exists(tmpdir):
            os.removedirs(tmpdir)


class InterrogateModels:
    blip_model = None
    clip_model = None
    clip_preprocess = None
    dtype = None
    running_on_cpu = None

    def __init__(self, content_dir):
        self.loaded_categories = None
        self.skip_categories = []
        self.content_dir = content_dir
        self.running_on_cpu = devices.device_interrogate == torch.device("cpu")

    def categories(self):
        if not os.path.exists(self.content_dir):
            download_default_clip_interrogate_categories(self.content_dir)

        if self.loaded_categories is not None and self.skip_categories == shared.opts.interrogate_clip_skip_categories:
           return self.loaded_categories

        self.loaded_categories = []

        if os.path.exists(self.content_dir):
            self.skip_categories = shared.opts.interrogate_clip_skip_categories
            category_types = []
            for filename in Path(self.content_dir).glob('*.txt'):
                category_types.append(filename.stem)
                if filename.stem in self.skip_categories:
                    continue
                m = re_topn.search(filename.stem)
                topn = 1 if m is None else int(m.group(1))
                with open(filename, "r", encoding="utf8") as file:
                    lines = [x.strip() for x in file.readlines()]

                self.loaded_categories.append(Category(name=filename.stem, topn=topn, items=lines))

        return self.loaded_categories

    def create_fake_fairscale(self):
        class FakeFairscale:
            def checkpoint_wrapper(self):
                pass

        sys.modules["fairscale.nn.checkpoint.checkpoint_activations"] = FakeFairscale

    def load_blip_model(self):
        self.create_fake_fairscale()
        import models.blip

        files = modelloader.load_models(
            model_path=os.path.join(paths.models_path, "BLIP"),
            model_url='https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_caption_capfilt_large.pth',
            ext_filter=[".pth"],
            download_name='model_base_caption_capfilt_large.pth',
        )

        blip_model = models.blip.blip_decoder(pretrained=files[0], image_size=blip_image_eval_size, vit='base', med_config=os.path.join(paths.paths["BLIP"], "configs", "med_config.json"))
        blip_model.eval()

        return blip_model

    def load_clip_model(self):
        import clip

        if self.running_on_cpu:
            model, preprocess = clip.load(clip_model_name, device="cpu", download_root=shared.cmd_opts.clip_models_path)
        else:
            model, preprocess = clip.load(clip_model_name, download_root=shared.cmd_opts.clip_models_path)

        model.eval()
        model = model.to(devices.device_interrogate)

        return model, preprocess

    def load(self):
        if self.blip_model is None:
            self.blip_model = self.load_blip_model()
            if not shared.cmd_opts.no_half and not self.running_on_cpu:
                self.blip_model = self.blip_model.half()

        self.blip_model = self.blip_model.to(devices.device_interrogate)

        if self.clip_model is None:
            self.clip_model, self.clip_preprocess = self.load_clip_model()
            if not shared.cmd_opts.no_half and not self.running_on_cpu:
                self.clip_model = self.clip_model.half()

        self.clip_model = self.clip_model.to(devices.device_interrogate)

        self.dtype = torch_utils.get_param(self.clip_model).dtype

    def send_clip_to_ram(self):
        if not shared.opts.interrogate_keep_models_in_memory:
            if self.clip_model is not None:
                self.clip_model = self.clip_model.to(devices.cpu)

    def send_blip_to_ram(self):
        if not shared.opts.interrogate_keep_models_in_memory:
            if self.blip_model is not None:
                self.blip_model = self.blip_model.to(devices.cpu)

    def unload(self):
        self.send_clip_to_ram()
        self.send_blip_to_ram()

        devices.torch_gc()

    def rank(self, image_features, text_array, top_count=1):
        import clip

        devices.torch_gc()

        if shared.opts.interrogate_clip_dict_limit != 0:
            text_array = text_array[0:int(shared.opts.interrogate_clip_dict_limit)]

        top_count = min(top_count, len(text_array))
        text_tokens = clip.tokenize(list(text_array), truncate=True).to(devices.device_interrogate)
        text_features = self.clip_model.encode_text(text_tokens).type(self.dtype)
        text_features /= text_features.norm(dim=-1, keepdim=True)

        similarity = torch.zeros((1, len(text_array))).to(devices.device_interrogate)
        for i in range(image_features.shape[0]):
            similarity += (100.0 * image_features[i].unsqueeze(0) @ text_features.T).softmax(dim=-1)
        similarity /= image_features.shape[0]

        top_probs, top_labels = similarity.cpu().topk(top_count, dim=-1)
        return [(text_array[top_labels[0][i].numpy()], (top_probs[0][i].numpy()*100)) for i in range(top_count)]

    def generate_caption(self, pil_image):
        gpu_image = transforms.Compose([
            transforms.Resize((blip_image_eval_size, blip_image_eval_size), interpolation=InterpolationMode.BICUBIC),
            transforms.ToTensor(),
            transforms.Normalize((0.48145466, 0.4578275, 0.40821073), (0.26862954, 0.26130258, 0.27577711))
        ])(pil_image).unsqueeze(0).type(self.dtype).to(devices.device_interrogate)

        with torch.no_grad():
            caption = self.blip_model.generate(gpu_image, sample=False, num_beams=shared.opts.interrogate_clip_num_beams, min_length=shared.opts.interrogate_clip_min_length, max_length=shared.opts.interrogate_clip_max_length)

        return caption[0]

    def interrogate(self, pil_image):
        res = ""
        shared.state.begin(job="interrogate")
        try:
            lowvram.send_everything_to_cpu()
            devices.torch_gc()

            self.load()

            caption = self.generate_caption(pil_image)
            self.send_blip_to_ram()
            devices.torch_gc()

            res = caption

            clip_image = self.clip_preprocess(pil_image).unsqueeze(0).type(self.dtype).to(devices.device_interrogate)

            with torch.no_grad(), devices.autocast():
                image_features = self.clip_model.encode_image(clip_image).type(self.dtype)

                image_features /= image_features.norm(dim=-1, keepdim=True)

                for cat in self.categories():
                    matches = self.rank(image_features, cat.items, top_count=cat.topn)
                    for match, score in matches:
                        if shared.opts.interrogate_return_ranks:
                            res += f", ({match}:{score/100:.3f})"
                        else:
                            res += f", {match}"

        except Exception:
            errors.report("Error interrogating", exc_info=True)
            res += "<error>"

        self.unload()
        shared.state.end()

        return res