fix image ratio reduction

examples/reachy2/demo-dev.yml

@@ -34,7 +34,7 @@ nodes:
     outputs:
       - response_base
     env:
-      ROBOT_IP: 10.42.0.80
+      ROBOT_IP: 172.17.134.85
 
   - id: reachy-left-arm
     build: pip install -e ../../node-hub/dora-reachy2
@@ -44,7 +44,7 @@ nodes:
     outputs:
       - response_l_arm
     env:
-      ROBOT_IP: 10.42.0.80
+      ROBOT_IP: 172.17.134.85
 
   - id: reachy-right-arm
     build: pip install -e ../../node-hub/dora-reachy2
@@ -54,7 +54,7 @@ nodes:
     outputs:
       - response_r_arm
     env:
-      ROBOT_IP: 10.42.0.80
+      ROBOT_IP: 172.17.134.85
 
   - id: reachy-camera
     build: pip install -e ../../node-hub/dora-reachy2
@@ -67,7 +67,7 @@ nodes:
       - image_depth
       - depth
     env:
-      ROBOT_IP: 10.42.0.80
+      ROBOT_IP: 172.17.134.85
 
   - id: reachy-head
     build: pip install -e ../../node-hub/dora-reachy2
@@ -76,20 +76,20 @@ nodes:
       boxes2d: parse_bbox/bbox_face
       look: state_machine/look
     env:
-      ROBOT_IP: 10.42.0.80
+      ROBOT_IP: 172.17.134.85
 
-  - id: plot
-    build: pip install -e ../../node-hub/dora-rerun
-    path: dora-rerun
-    inputs:
-      camera_left/image_right: reachy-camera/image_right
-      camera_torso/image: reachy-camera/image_depth
-      text_response: dora-qwenvl/text
-      text_whisper: dora-distil-whisper/text
-      camera_torso/boxes2d: parse_bbox/bbox
-      camera_left/boxes2d_face: parse_bbox/bbox_face
-    env:
-      RERUN_MEMORY_LIMIT: 5%
+  #- id: plot
+  #build: pip install -e ../../node-hub/dora-rerun
+  #path: dora-rerun
+  #inputs:
+  #camera_left/image_right: reachy-camera/image_right
+  #camera_torso/image: reachy-camera/image_depth
+  #text_response: dora-qwenvl/text
+  #text_whisper: dora-distil-whisper/text
+  #camera_torso/boxes2d: parse_bbox/bbox
+  #camera_left/boxes2d_face: parse_bbox/bbox_face
+  #env:
+  #RERUN_MEMORY_LIMIT: 5%
 
   - id: dora-qwenvl
     build: pip install -e ../../node-hub/dora-qwen2-5-vl
@@ -104,6 +104,7 @@ nodes:
     env:
       #ADAPTER_PATH: /home/peter/Documents/work/LLaMA-Factory/saves/qwen2.5_vl-felix/lora/sft/checkpoint-558
       DEFAULT_QUESTION: grab human.
+      IMAGE_RESIZE_RATIO: "0.5"
       # ACTIVATION_WORDS: grab pick give output take catch grabs picks gives output takes catches have
       #SYSTEM_PROMPT: You're a robot.
 
@@ -115,6 +116,8 @@ nodes:
       - bbox
       - bbox_face
       - action_arm
+    env:
+      IMAGE_RESIZE_RATIO: "0.5"
 
   - id: box_coordinates
     build: pip install -e ../../node-hub/dora-boxes2d-to-pose

examples/reachy2/parse_bbox.py

@@ -1,4 +1,5 @@
 import json
+import os
 
 import numpy as np
 import pyarrow as pa
@@ -6,6 +7,8 @@ from dora import Node
 
 node = Node()
 
+IMAGE_RESIZE_RATIO = float(os.getenv("IMAGE_RESIZE_RATIO", "1.0"))
+
 
 def extract_bboxes(json_text):
     """
@@ -45,7 +48,7 @@ for event in node:
 
         bboxes, labels = extract_bboxes(text)
         if bboxes is not None and len(bboxes) > 0:
-            bboxes = bboxes  # * 2
+            bboxes = bboxes * int(1 / IMAGE_RESIZE_RATIO)
             if "human" in labels[0] or "head" in labels[0]:
                 node.send_output("bbox_face", pa.array(bboxes.ravel()))
             else:

examples/reachy2/state_machine.py

@@ -1,4 +1,4 @@
-## State Machine
+# State Machine
 import os
 import time
 
@@ -105,18 +105,17 @@ def wait_for_events(ids: list[str]):
 
 
 while True:
-
     ### === IDLE ===
 
     node.send_output(
         "action_r_arm",
         pa.array(r_default_pose),
-        metadata={"encoding": "jointstate"},
+        metadata={"encoding": "jointstate", "duration": 2},
     )
     node.send_output(
         "action_l_arm",
         pa.array(l_default_pose),
-        metadata={"encoding": "jointstate"},
+        metadata={"encoding": "jointstate", "duration": 2},
     )
     wait_for_events(ids=["response_r_arm", "response_l_arm"])
 
@@ -182,7 +181,9 @@ while True:
         y = values[1]
         z = values[2]
         x = x + 0.04
-        z = np.clip(z, -0.31, -0.22)
+
+        ## Clip the Maximum and minim values for the height of the arm to avoid collision or weird movement.
+        z = np.clip(z, -0.32, -0.22)
         node.send_output("look", pa.array([x, y, z]))
         trajectory = np.array(
             [

node-hub/dora-qwen2-5-vl/dora_qwen2_5_vl/main.py

@@ -72,8 +72,8 @@ def generate(frames: dict, question, history, past_key_values=None, select_image
                 {
                     "type": "image",
                     "image": image,
-                    "resized_height": image.size[1] / IMAGE_RESIZE_RATIO,
-                    "resized_width": image.size[0] / IMAGE_RESIZE_RATIO,
+                    "resized_height": image.size[1] * IMAGE_RESIZE_RATIO,
+                    "resized_width": image.size[0] * IMAGE_RESIZE_RATIO,
                 }
                 for image in images
             ]