fix(api): convert Real ESRGAN v3 using same arch as runtime

ssube · Feb 18, 2023 · 338fc23 · 338fc23
1 parent 2c9d96d
commit 338fc23
Show file tree

Hide file tree

Showing 2 changed files with 28 additions and 13 deletions.
diff --git a/api/onnx_web/chain/upscale_resrgan.py b/api/onnx_web/chain/upscale_resrgan.py
@@ -14,7 +14,7 @@
 last_pipeline_instance = None
 last_pipeline_params = (None, None)
 
-x4_v3_tag = "real-esrgan-x4-v3"
+TAG_X4_V3 = "real-esrgan-x4-v3"
 
 
 def load_resrgan(
@@ -37,7 +37,7 @@ def load_resrgan(
     if not path.isfile(model_path):
         raise Exception("Real ESRGAN model not found at %s" % model_path)
 
-    elif params.format == "onnx":
+    if params.format == "onnx":
         # use ONNX acceleration, if available
         model = OnnxNet(
             server,
@@ -46,7 +46,7 @@ def load_resrgan(
             sess_options=device.sess_options(),
         )
     elif params.format == "pth":
-        if x4_v3_tag in model_file:
+        if TAG_X4_V3  in model_file:
             # the x4-v3 model needs a different network
             model = SRVGGNetCompact(
                 num_in_ch=3,
@@ -69,8 +69,8 @@ def load_resrgan(
         raise Exception("unknown platform %s" % params.format)
 
     dni_weight = None
-    if params.upscale_model == x4_v3_tag and params.denoise != 1:
-        wdn_model_path = model_path.replace(x4_v3_tag, "%s-wdn" % (x4_v3_tag))
+    if params.upscale_model == TAG_X4_V3 and params.denoise != 1:
+        wdn_model_path = model_path.replace(TAG_X4_V3, "%s-wdn" % TAG_X4_V3)
         model_path = [model_path, wdn_model_path]
         dni_weight = [params.denoise, 1 - params.denoise]
 

diff --git a/api/onnx_web/convert/upscale_resrgan.py b/api/onnx_web/convert/upscale_resrgan.py
@@ -3,12 +3,15 @@
 
 import torch
 from basicsr.archs.rrdbnet_arch import RRDBNet
+from realesrgan.archs.srvgg_arch import SRVGGNetCompact
 from torch.onnx import export
 
 from .utils import ConversionContext, ModelDict
 
 logger = getLogger(__name__)
 
+TAG_X4_V3 = "real-esrgan-x4-v3"
+
 
 @torch.no_grad()
 def convert_upscale_resrgan(
@@ -28,14 +31,26 @@ def convert_upscale_resrgan(
         return
 
     logger.info("loading and training model")
-    model = RRDBNet(
-        num_in_ch=3,
-        num_out_ch=3,
-        num_feat=64,
-        num_block=23,
-        num_grow_ch=32,
-        scale=scale,
-    )
+
+    if TAG_X4_V3 in name:
+        # the x4-v3 model needs a different network
+        model = SRVGGNetCompact(
+            num_in_ch=3,
+            num_out_ch=3,
+            num_feat=64,
+            num_conv=32,
+            upscale=scale,
+            act_type="prelu",
+        )
+    else:
+        model = RRDBNet(
+            num_in_ch=3,
+            num_out_ch=3,
+            num_feat=64,
+            num_block=23,
+            num_grow_ch=32,
+            scale=scale,
+        )
 
     torch_model = torch.load(source, map_location=ctx.map_location)
     if "params_ema" in torch_model: