Failed to get any callback frames to render
ElinLiu0 opened this issue · 1 comments
ElinLiu0 commented
Hi,i'm trying to use NVIDIA Triton Inference Server and webRTC to build a detection app.
I have write the recv()
function in VideoProcessor like below:
def recv(self,frame):
inputs,outputs = [],[]
frame = frame.to_ndarray(format="bgr24")
image_data = self.preprocess(frame)
inputs.append(httpclient.InferInput(self.inputName,image_data.shape,"FP32"))
inputs[0].set_data_from_numpy(image_data)
outputs.append(httpclient.InferRequestedOutput(self.outputName))
try:
results = self.inferenceClient.async_infer(
model_name = self.modelName,
inputs = inputs,
outputs = outputs,
model_version = self.modelVersion,
request_id = str(random.random())
)
output = results.get_result().as_numpy(self.outputName)
outputFrame = self.postprocess(frame,output)
return av.VideoFrame.from_ndarray(outputFrame,format="bgr24")
except InferenceServerException as e:
return str(e)
Yet it doesn't render the callback frame on my app:
But i can prove that inference Server working fine,with the same code on opencv local reading script,it works fine:
So is there any way to deal with that?
ElinLiu0 commented
Closed,this issue were Due to the different shooting angles in trainig set,then yolo model can't detect the object which front of camera