Hi, I trained resnet_18 two class classifier. I get accuracy around 98%, when I infer using tlt-infer, classification are perfect with 90% confidence. But when I converted it to etlt->engine and used it in deepstream, for same frame I get different output in deepstream which incorrectly classifies,…

There is no update from you for a period, assuming this is not an issue any more. Hence we are closing this topic. If need further support, please open a new one. Thanks Please paste your running command. And is it possible for you to attach a screenshot for the wrong inference?

Hi sathiez, Which data type did you tlt-export, int8 or fp16? More, could you please paste the config file and the command line along with logs when you run deepsteam? Thanks.

Hi, I am using fp32 [property] gpu-id=0 net-scale-factor=1 model-engine-file=classify.engine batch-size=2 # 0=FP32 and 1=INT8 mode network-mode=0 process-mode=2 model-color-format=0 gpu-id=0 gie-unique-id=3 operate-on-gie-id=1 operate-on-class-ids=0 output-blob-names=predictions/Softmax #offsets …

Can you paste your command line too? Thanks.

Can you refer to “Integrating a Classification model” part of tlt user guide Integrating TAO Models into DeepStream — TAO Toolkit 3.22.05 documentation ? In your config file, there is not “labelfile-path”, “input-dims”, “uff-input-blob-name”.

Hi, I am using deepstream-infer-tensor-meta-app so I will mention labels inside .cpp file. Even after I try using “input-dims”, “uff-input-blob-name” my output remains same. my command line ./deepstream-infer-tensor-meta-app video.h264

I already tried with deepstream-test4 app I get same results.

How many frames in your h264 files? What’s the accuracy rate when you run deepstream?Do you mean you generate the h264 file with the same images as using in tlt-infer?

Hi sathiez, Have you managed to get issue resolved? Any result can be shared?

For same frame I get different output using .tlt and .engine

Accelerated Computing Intelligent Video Analytics TAO Toolkit

Morganh December 14, 2019, 5:14am 7

Can you use deepstream_test1_app to check too? Thanks.

Topic		Replies	Views
Tlt resnet18 performance drop between .tlt inference and .engine TAO Toolkit	24	2544	August 5, 2021
Inferring resnet18 classification etlt model with python TAO Toolkit	44	4787	February 11, 2021
Issue with image classification tutorial and testing with deepstream-app TAO Toolkit tensorrt , jetson-inference	33	6534	May 25, 2021
Different class label between tlt-infer and generated TensorRT engine on DeepStream TAO Toolkit	7	651	July 30, 2020
Tlt3.0 train yolov4 of resnet10, "tlt yolo_v4 inference" could get right bboxes, but deepstream5.1 get wrong result TAO Toolkit	8	829	April 15, 2021
Different outputs for tlt and trt TAO Toolkit	6	679	December 29, 2021
Error running ResNet18 classifier in deepstream TAO Toolkit	1	609	February 12, 2021
Issues with tao classifier_tf2 in deepstream (Accuracy drops) TAO Toolkit deepstream	20	507	September 6, 2024
Issue running tlt trained SSD-resnet18 on Xavier with deepstream-app TAO Toolkit	14	941	September 21, 2020
How to have more accurate classification results in deepstream? TAO Toolkit	24	1851	February 4, 2020

For same frame I get different output using .tlt and .engine

Related topics