feat(api): add CodeFormer to automatic upscale

api/onnx_web/chain/correct_codeformer.py

@@ -9,21 +9,22 @@
 
 logger = getLogger(__name__)
 
-pretrain_model_url = (
-    "https://github.com/sczhou/CodeFormer/releases/download/v0.1.0/codeformer.pth"
-)
-
 device = "cpu"
 
 
 def correct_codeformer(
-    _job: JobContext,
+    job: JobContext,
     _server: ServerContext,
     _stage: StageParams,
     _params: ImageParams,
-    source_image: Image.Image,
+    source: Image.Image,
+    *,
+    source_image: Image.Image = None,
     **kwargs,
 ) -> Image.Image:
-    pipe = CodeFormer().to(device)
+    device = job.get_device()
+    # TODO: terrible names, fix
+    image = source or source_image
 
-    return pipe(source_image)
+    pipe = CodeFormer().to(device.torch_device())
+    return pipe(image)

api/onnx_web/upscale.py

@@ -4,6 +4,7 @@
 
 from .chain import (
     ChainPipeline,
+    correct_codeformer,
     correct_gfpgan,
     upscale_resrgan,
     upscale_stable_diffusion,
@@ -40,9 +41,16 @@ def run_upscale_correction(
             mini_tile = min(SizeChart.mini, stage.tile_size)
             stage = StageParams(tile_size=mini_tile, outscale=upscale.outscale)
             chain.append((upscale_stable_diffusion, stage, None))
+        else:
+            logger.warn("unknown upscaling model: %s", upscale.upscale_model)
 
     if upscale.faces:
         stage = StageParams(tile_size=stage.tile_size, outscale=1)
-        chain.append((correct_gfpgan, stage, None))
+        if "codeformer" in upscale.correction_model:
+            chain.append((correct_codeformer, stage, None))
+        elif "gfpgan" in upscale.correction_model:
+            chain.append((correct_gfpgan, stage, None))
+        else:
+            logger.warn("unknown correction model: %s", upscale.correction_model)
 
     return chain(job, server, params, image, prompt=params.prompt, upscale=upscale)