From 624b926e336c5cff572d7642dddd9a45eae60770 Mon Sep 17 00:00:00 2001
From: gouhanke <12219217+gouhanke@user.noreply.gitee.com>
Date: Thu, 12 Feb 2026 17:14:23 +0800
Subject: [PATCH] =?UTF-8?q?debug:=20=E6=B7=BB=E5=8A=A0=E6=8E=A8=E7=90=86?=
 =?UTF-8?q?=E6=97=B6=E7=BC=A9=E6=94=BE=EF=BC=8C=E5=8A=A0=E5=A4=A7=E9=87=87?=
 =?UTF-8?q?=E6=95=B0=E4=BB=A5=E5=8F=8A=E6=8E=A8=E7=90=86=E6=97=B6=E7=89=A9?=
 =?UTF-8?q?=E5=9D=97=E7=9A=84=E6=94=BE=E7=BD=AE=E8=8C=83=E5=9B=B4?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 roboimi/demos/vla_scripts/eval_vla.py | 4 ++++
 roboimi/utils/act_ex_utils.py         | 4 ++--
 2 files changed, 6 insertions(+), 2 deletions(-)

diff --git a/roboimi/demos/vla_scripts/eval_vla.py b/roboimi/demos/vla_scripts/eval_vla.py
index 97fe38f..9c358e4 100644
--- a/roboimi/demos/vla_scripts/eval_vla.py
+++ b/roboimi/demos/vla_scripts/eval_vla.py
@@ -103,10 +103,14 @@ def prepare_observation(obs: Dict, camera_names: list) -> Dict:
     Returns:
         agent 格式的观测字典
     """
+    import cv2
+
     # 转换图像: numpy -> tensor, HWC -> CHW
     images = {}
     for cam_name in camera_names:
         img = obs['images'][cam_name]
+        # Resize 到 224x224（与训练时一致）
+        img = cv2.resize(img, (224, 224), interpolation=cv2.INTER_LINEAR)
         img = rearrange(img, 'h w c -> c h w')
         img = torch.from_numpy(img / 255.0).float()
         images[cam_name] = img
diff --git a/roboimi/utils/act_ex_utils.py b/roboimi/utils/act_ex_utils.py
index 3c1648e..d08f203 100644
--- a/roboimi/utils/act_ex_utils.py
+++ b/roboimi/utils/act_ex_utils.py
@@ -27,8 +27,8 @@ def sample_insertion_pose():
 
 def sample_transfer_pose():
         # Box
-        x_range = [0.0, 0.05]
-        y_range = [0.95, 1.05]
+        x_range = [-0.05, 0.05]
+        y_range = [0.90, 1.05]
         z_range = [0.47, 0.47]
 
         ranges = np.vstack([x_range, y_range, z_range])