Update demo.py

demo.py

@@ -18,7 +18,7 @@ model = AutoModelForCausalLM.from_pretrained(
     trust_remote_code=True,
     torch_dtype=torch.float16,
     device_map={"": "cuda"},
-    revision="6357578ba3bf29aad59900d22751102d175f4976"
+    revision="56a3adeae60809e4269c544cde376feb20637ee0"
 )
 
 
@@ -103,6 +103,8 @@ def process_image(input_image):
 
         # Get image encoding
         enc_image = model.encode_image(pil_image)
+        flipped_pil = image.copy().transpose(method=Image.FLIP_LEFT_RIGHT)
+        flip_enc_image = self.encode_image(flipped_pil)
 
         # Detect faces
         faces = model.detect(enc_image, "face")["objects"]
@@ -115,7 +117,10 @@ def process_image(input_image):
         gaze_points = []
 
         for face in faces:
-            gaze = model.detect_gaze(pil_image, face=face)["gaze"]
+            gaze = model.detect_gaze(enc_image, face=face, unstable_settings={
+                "prioritize_accuracy": True,
+                "flip_enc_image": flip_enc_image
+            })["gaze"]
 
             if gaze is None:
                 continue