Deeplabcut Live for Multi-Animal #130

I have used Deeplabcut GUI to train a model for poses of two rats, which includes three multi-animal-bodyparts and two unique-bodyparts. The model was tested in Deeplab GUI through a video and the result shows that all the 8 points could be dotted.
The main purpose of us is for real-time application, so we find Deeplabcut-Live for next step. We import the model through python code:
deeplabcut.export_model(config_path, shuffle=1, make_tar=False)
Then a new folder named exported-models was created, which included a configuration file pose_cfg.yaml and other model files. Finally, I used deeplabcut-live to process a test video in real time, code:

import cv2
from dlclive import DLCLive, Processor
import time

model_path = r"C:\Users\XHB\Desktop\Multi_Rat-HB-2025-07-31\exported-models\DLC_Multi_Rat_resnet_50_iteration-0_shuffle-1"
dlc_proc = Processor()
dlc_live = DLCLive(model_path, processor=dlc_proc)
cap = cv2.VideoCapture("2rat2.avi")
ret, frame = cap.read()
dlc_live.init_inference(frame)

while True:
    if not ret:
        break
    pose = dlc_live.get_pose(frame)
    time_end = time.time()
    for x, y, p in pose:
        if p > 0.5: 
            cv2.circle(frame, (int(x), int(y)), 5, (0, 255, 0), -1)
    cv2.imshow("DLC Live", frame)
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

cap.release()

The result shows that only 5 points could be dotted, including the 3 multi-animal-bodyparts of the first rat and the 2 unique-bodyparts, while the other 3 multi-animal-bodyparts of the second rat were neglected.
Thus, I wanna ask for help, how could I recognize all the 8 parts of the 2 rats.

result in Deeplabcut GUI:

result in Deeplabcut-Live:

configuration file in Deeplabcut GUI:
config.zip

configuration file in Deeplabcut GUI:
pose_cfg.zip