Open iMbaGong opened 3 days ago
@iMbaGong Can you provide examples for the output scores of 100%. Is confidence only getting wrong or the boxes are also issue?
Also, The operator einsum
might cause issues with older 23.01
https://github.com/AILab-CVC/YOLO-World/blob/master/docs/deploy.md#faq
just like this
And the operator einsum
didn't cause a error in 23.01
@PrinceP You can reproduce the problem by docker below
FROM nvcr.io/nvidia/cuda:11.4.3-devel-ubuntu20.04
# Install required packages for building and running the sample project
RUN apt-get -y update \
&& DEBIAN_FRONTEND=noninteractive apt-get install -y tzdata \
&& apt-get update && apt-get install -y \
build-essential \
cmake \
git \
pkg-config \
libavcodec-dev \
libhdf5-dev \
libavformat-dev \
libavdevice-dev \
libcurl4-openssl-dev \
libssl-dev \
wget
RUN apt-get install -y \
libopencv-dev
ENV cuda_version cuda11.8
ENV cudnn_version 8.6.0.163
ENV trt_version 8.5.2
RUN apt-get update \
&& apt-get install libcudnn8=${cudnn_version}-1+${cuda_version} \
libcudnn8-dev=${cudnn_version}-1+${cuda_version} \
libnvinfer8=${trt_version}-1+${cuda_version} \
libnvonnxparsers8=${trt_version}-1+${cuda_version} \
libnvparsers8=${trt_version}-1+${cuda_version} \
libnvinfer-dev=${trt_version}-1+${cuda_version} \
libnvonnxparsers-dev=${trt_version}-1+${cuda_version} \
libnvparsers-dev=${trt_version}-1+${cuda_version}
ENV SPDLOGGER_VERSION 1.12.0
RUN wget -P /tmp https://github.com/gabime/spdlog/archive/v${SPDLOGGER_VERSION}.tar.gz
RUN tar -C /tmp -xzf /tmp/v${SPDLOGGER_VERSION}.tar.gz
RUN cp -R /tmp/spdlog-${SPDLOGGER_VERSION}/include/spdlog /usr/include/
RUN rm /tmp/v${SPDLOGGER_VERSION}.tar.gz
RUN rm -rf /workspace
WORKDIR /app
ENV NVIDIA_VISIBLE_DEVICES=all
ENV OPENCV_FFMPEG_LOGLEVEL=0
CMD [ "sleep", "infinity"]
Hi, thank for great work! When I try to build and run the yolo-world tensorrt model on nvidia orin nx(with cuda11.4 and tensorrt8.5.2), i meet a mistake. It's output score all became 100%. When i test on host machine, when docker build from
nvcr.io/nvidia/tensorrt:24.02-py3
, it is ok. But when it build fromnvcr.io/nvidia/tensorrt:23.01-py3
, the same problem happened.