Run RTMLib benchmark outside production startup

servers/unraid/tail2-pose-server/compose.yaml

@@ -166,6 +166,38 @@ services:
         } > "$$out/process-map.csv"
         chmod -R a+rX,u+w "$$out"
 
+  benchmark-rtmlib-body:
+    image: tail2-pose-server:0.1.0
+    container_name: tail2-pose-benchmark-rtmlib-body
+    gpus: all
+    restart: "no"
+    entrypoint: ["/bin/bash", "-lc"]
+    depends_on:
+      models:
+        condition: service_completed_successfully
+    volumes:
+      - /mnt/user/appdata/tail2-pose-server/app:/app:ro
+      - /mnt/user/appdata/tail2-pose-server/models:/models:ro
+      - ./validation:/validation
+    environment:
+      PYTHONPATH: /app
+      CUDA_VISIBLE_DEVICES: "1"
+      MODEL_BACKEND: rtmlib_body
+      BENCHMARK_PROFILE: rtmlib_body_yolox-m_rtmpose-m_gpu1_threads2
+      BENCHMARK_OUTPUT: /validation/benchmark-rtmlib-body.json
+      DETECTION_THRESHOLD: "0.35"
+      NMS_IOU_THRESHOLD: "0.60"
+      MAX_PEOPLE: "4"
+      DET_CHECKPOINT: /models/yolox-m-humanart.onnx
+      POSE_CHECKPOINT: /models/rtmpose-m-body17.onnx
+      OMP_NUM_THREADS: "2"
+      MKL_NUM_THREADS: "2"
+      OPENBLAS_NUM_THREADS: "2"
+      CV2_NUM_THREADS: "2"
+    command:
+      - |
+        python -c 'from tail2.models import Models; from tail2.offline_benchmark import run_offline_benchmark; m=Models(); m.load(); run_offline_benchmark(m, .35, 4)'
+
   validate:
     profiles: ["validation"]
     image: tail2-pose-server:0.1.0