Error "pycuda._driver.LogicError: cuMemcpyHtoD failed: invalid device context" While using Flask and TensorRT engine

Description

Hi
I’m using a TensorRT engine to infer batch images that are received from flask request. i succesfully build engine using infer.py > TensorRTInfer function. you can see the code in bellow link

after that when i received my images i use ImageBatcher to get appropriate batches to inference TensorRT engine.
My code is as bellow
batcher = ImageBatcher(image_list, TRTengine.input_spec())
** for batch, images, scales in batcher.get_batch():
*
** print(“Processing Image {} / {}”.format(batcher.image_index, batcher.num_images), end=“\r”)**
** detections = TRTengine.infer(batch, scales, nms_threshold=0.5)**

while my TRTengine.infer function is as bellow:

def infer(self, batch, scales=None, nms_threshold=None):
** outputs = **
** for shape, dtype in self.output_spec():**
** outputs.append(np.zeros(shape, dtype))**
** cuda.memcpy_htod(self.inputs[0][‘allocation’], np.ascontiguousarray(batch))**

but i receive the following error:
File “/home/user/Desktop/python/projects/app.py”, line 195, in object_detection
detections = TRTengine.infer(batch, scales, nms_threshold=0.5)
File “/home/user/Desktop/python/projects/infer.py”, line 116, in infer
cuda.memcpy_htod(self.inputs[0][‘allocation’], np.ascontiguousarray(batch))
pycuda._driver.LogicError: cuMemcpyHtoD failed: invalid device context

whats the problem?

Environment

TensorRT Version: 8.0.3
GPU Type: RTX 2080 Ti
Nvidia Driver Version: 470.57.02
CUDA Version: 11.3
CUDNN Version: –
Operating System + Version: Ubuntu 18.04
Python Version (if applicable): 3.6.9

Hi @anajar2198
Sorry, i am little confused. Are you using below code for inference or you made any custom changes as well? TensorRT/infer.py at main · NVIDIA/TensorRT · GitHub

In case you are using any modified code version, could you please share the code and model so we can help better?

Thanks

Hi
I’m using the bellow link to do TRT inference, and i didn’t do any modification in the TRT inference.
I use the same functions but when I’m using memcpy_htod function, I receive this weird error, which hasn’t been mentioned in other cites or foroms.
thanks