[Safety Checker] Add Safety Checker Module

Former-commit-id: d0c714ae4afa1c011269a956d6f260f84f77025e

scripts/txt2img.py

@@ -16,12 +16,29 @@ from ldm.util import instantiate_from_config
 from ldm.models.diffusion.ddim import DDIMSampler
 from ldm.models.diffusion.plms import PLMSSampler
 
+from diffusers.pipelines.stable_diffusion.safety_checker import StableDiffusionSafetyChecker
+from transformers import AutoFeatureExtractor
+
+feature_extractor = AutoFeatureExtractor.from_pretrained("CompVis/stable-diffusion-v-1-3", use_auth_token=True)
+safety_checker = StableDiffusionSafetyChecker.from_pretrained("CompVis/stable-diffusion-v-1-3", use_auth_token=True)
 
 def chunk(it, size):
     it = iter(it)
     return iter(lambda: tuple(islice(it, size)), ())
 
 
+def numpy_to_pil(images):
+    """
+    Convert a numpy image or a batch of images to a PIL image.
+    """
+    if images.ndim == 3:
+        images = images[None, ...]
+    images = (images * 255).round().astype("uint8")
+    pil_images = [Image.fromarray(image) for image in images]
+
+    return pil_images
+
+
 def load_model_from_config(config, ckpt, verbose=False):
     print(f"Loading model from {ckpt}")
     pl_sd = torch.load(ckpt, map_location="cpu")
@@ -220,7 +237,9 @@ def main():
     if opt.fixed_code:
         start_code = torch.randn([opt.n_samples, opt.C, opt.H // opt.f, opt.W // opt.f], device=device)
 
+    print("start code", start_code.abs().sum())
     precision_scope = autocast if opt.precision=="autocast" else nullcontext
+    precision_scope = nullcontext
     with torch.no_grad():
         with precision_scope("cuda"):
             with model.ema_scope():
@@ -269,7 +288,11 @@ def main():
                     Image.fromarray(grid.astype(np.uint8)).save(os.path.join(outpath, f'grid-{grid_count:04}.png'))
                     grid_count += 1
 
-                toc = time.time()
+                image = x_samples_ddim.cpu().permute(0, 2, 3, 1).numpy()
+
+                # run safety checker
+                safety_checker_input = pipe.feature_extractor(numpy_to_pil(image), return_tensors="pt")
+                image, has_nsfw_concept = pipe.safety_checker(images=image, clip_input=safety_checker_input.pixel_values)
 
     print(f"Your samples are ready and waiting for you here: \n{outpath} \n"
           f" \nEnjoy.")