update

node_scripts/deva_node.py

@@ -3,6 +3,7 @@
 
 import numpy as np
 import torch
+import torch.nn.functional as F
 import supervision as sv
 
 import rospy
@@ -14,7 +15,7 @@
 from jsk_recognition_msgs.msg import ClassificationResult
 from jsk_recognition_msgs.msg import Label, LabelArray
 
-from deva.inference.demo_utils import get_input_frame_for_deva
+from deva.dataset.utils import im_normalization
 from deva.ext.grounding_dino import segment_with_text
 
 from model_config import SAMConfig, GroundingDINOConfig, DEVAConfig
@@ -73,16 +74,17 @@ def publish_result(self, mask, vis, frame_id):
     def callback(self, img_msg):
         self.image = self.bridge.imgmsg_to_cv2(img_msg, desired_encoding="rgb8")
         with torch.cuda.amp.autocast(enabled=self.cfg["amp"]):
-            min_size = min(self.image.shape[:2])
-            deva_input = get_input_frame_for_deva(self.image, min_size).to(self.deva_config.device)
+            h, w = self.image.shape[:2]
+            deva_input = im_normalization(torch.from_numpy(self.image).permute(2, 0, 1).float() / 255).unsqueeze(0).to(self.deva_config.device)
+            deva_input = F.interpolate(deva_input, (h, w), mode='bilinear', align_corners=False)[0]
             if self.cnt % self.cfg["detection_every"] == 0:
                 incorporate_mask, segments_info = segment_with_text(
                     self.cfg,
                     self.gd_predictor,
                     self.sam_predictor,
                     self.image,
                     self.classes,
-                    min_size,
+                    min(h, w),
                 )
                 prob = self.deva_predictor.incorporate_detection(deva_input, incorporate_mask, segments_info)
                 self.object_ids = [seg.id for seg in segments_info]