Merge branch 'main' into add_seg_weights

Author: PawelPeczek-Roboflow

.release/pypi/inference.cli.setup.py

@@ -65,6 +65,7 @@ def read_requirements(path):
     },
     extras_require={
         "cloud-deploy": read_requirements("requirements/requirements.cloud_deploy.txt"),
+        "cloud-storage": read_requirements("requirements/requirements.cloud_storage.txt"),
     },
     package_data={"": ["configs/*.yml"]},
     install_requires=read_requirements([

docs/install/jetson.md

@@ -36,7 +36,21 @@ If you want more control of the container settings you can also start it
 manually. Jetson devices with NVIDIA Jetpack are pre-configured with NVIDIA Container
 Runtime and will be hardware accelerated out of the box:
 
-=== "Jetpack 6"
+=== "Jetpack 6.2"
+    ```bash
+    sudo docker run -d \
+        --name inference-server \
+        --runtime nvidia \
+        --read-only \
+        -p 9001:9001 \
+        --volume ~/.inference/cache:/tmp:rw \
+        --security-opt="no-new-privileges" \
+        --cap-drop="ALL" \
+        --cap-add="NET_BIND_SERVICE" \
+        roboflow/roboflow-inference-server-jetson-6.2.0:latest
+    ```
+
+=== "Jetpack 6.0"
     ```bash
     sudo docker run -d \
         --name inference-server \
@@ -110,7 +124,22 @@ load each model.
 
 Enable TensorRT by adding `TensorrtExecutionProvider` to the `ONNXRUNTIME_EXECUTION_PROVIDERS` environment variable.
 
-=== "Jetpack 6"
+=== "Jetpack 6.2"
+    ```bash
+    sudo docker run -d \
+        --name inference-server \
+        --runtime nvidia \
+        --read-only \
+        -p 9001:9001 \
+        --volume ~/.inference/cache:/tmp:rw \
+        --security-opt="no-new-privileges" \
+        --cap-drop="ALL" \
+        --cap-add="NET_BIND_SERVICE" \
+        -e ONNXRUNTIME_EXECUTION_PROVIDERS="[TensorrtExecutionProvider,CUDAExecutionProvider,CPUExecutionProvider]" \
+        roboflow/roboflow-inference-server-jetson-6.2.0:latest
+    ```
+
+=== "Jetpack 6.0"
     ```bash
     sudo docker run -d \
         --name inference-server \
@@ -182,7 +211,37 @@ Enable TensorRT by adding `TensorrtExecutionProvider` to the `ONNXRUNTIME_EXECUT
 
 If you are using Docker Compose for your application, the equivalent yaml is:
 
-=== "Jetpack 6"
+=== "Jetpack 6.2"
+    ```yaml
+    version: "3.9"
+    
+    services:
+      inference-server:
+        container_name: inference-server
+        image: roboflow/roboflow-inference-server-jetson-6.2.0:latest
+    
+        read_only: true
+        ports:
+          - "9001:9001"
+
+        volumes:
+          - "${HOME}/.inference/cache:/tmp:rw"
+    
+        runtime: nvidia
+
+        # Optionally: uncomment the following lines to enable TensorRT:
+        # environment:
+        #   ONNXRUNTIME_EXECUTION_PROVIDERS: "[TensorrtExecutionProvider,CUDAExecutionProvider,CPUExecutionProvider]"
+    
+        security_opt:
+          - no-new-privileges
+        cap_drop:
+          - ALL
+        cap_add:
+          - NET_BIND_SERVICE
+    ```
+
+=== "Jetpack 6.0"
     ```yaml
     version: "3.9"
 

inference/core/interfaces/http/http_api.py

@@ -1,6 +1,7 @@
 import base64
 import concurrent
 import os
+import re
 from concurrent.futures import CancelledError, Future, ThreadPoolExecutor
 from functools import partial
 from threading import Lock, Thread
@@ -1468,7 +1469,14 @@ def validate_workflow(
             @with_route_exceptions_async
             async def initialise_webrtc_worker(
                 request: WebRTCWorkerRequest,
+                r: Request,
             ) -> InitializeWebRTCResponse:
+                if str(r.headers.get("origin")).lower() == BUILDER_ORIGIN.lower():
+                    if re.search(
+                        r"^https://[^.]+\.roboflow\.[^./]+/", str(r.url).lower()
+                    ):
+                        request.is_preview = True
+
                 logger.debug("Received initialise_webrtc_worker request")
                 worker_result: WebRTCWorkerResult = await start_worker(
                     webrtc_request=request,

inference/core/interfaces/webrtc_worker/entities.py

@@ -28,6 +28,7 @@ class WebRTCConfig(BaseModel):
 class WebRTCWorkerRequest(BaseModel):
     api_key: Optional[str] = None
     workflow_configuration: WorkflowConfiguration
+    is_preview: bool = False
     webrtc_offer: WebRTCOffer
     webrtc_config: Optional[WebRTCConfig] = None
     # TODO: to be removed, replaced with webrtc_config
@@ -39,9 +40,6 @@ class WebRTCWorkerRequest(BaseModel):
     data_output: Optional[List[str]] = Field(default=None)
     declared_fps: Optional[float] = None
     rtsp_url: Optional[str] = None
-    use_data_channel_frames: bool = (
-        False  # When True, expect frames via data channel instead of media track
-    )
     processing_timeout: Optional[int] = WEBRTC_MODAL_FUNCTION_TIME_LIMIT
     processing_session_started: Optional[datetime.datetime] = None
     requested_plan: Optional[str] = "webrtc-gpu-small"

inference/core/interfaces/webrtc_worker/modal.py

@@ -415,6 +415,7 @@ def send_answer(obj: WebRTCWorkerResult):
                     "plan": webrtc_plan,
                     "billable": True,
                     "video_source": video_source,
+                    "is_preview": webrtc_request.is_preview,
                 },
                 execution_duration=(
                     _exec_session_stopped - _exec_session_started