Low latency than expected#27
Open
paduck86 wants to merge 61 commits into
Open
Conversation
…about camera_tf_trt.py
Finished the first working implementation of the real-time object detection demo script: 'camera_tf_trt.py'
… drawing function
…onvActLayer.cpp (61) - Cuda Error in createFilterTextur'
…xes; fix duplicated logging for tensorflow
Refactor visualization code to the utils/ directory; fix duplicated logging for tensorflow
…th/height setting in command line arguments)
…rflow-1.8 (as specified in the original NVIDIA tf_trt_models/README.md)
…l ones from tensorflow model repository), in which the 'score_threshold' has been modified from 1e-8 to 0.3
Update a working version, tested with JetPack-3.2 and tensorflow 1.8.0
Add download link to tensorflow 1.8.0 wheel for JetPack-3.3
Add support for ssd_mobilenet_v1_egohands
Add data/egohands_label_map.pbtxt
…ohands', 'ssd_inception_v2_egohands', 'faster_rcnn_resnet50_egohands', 'faster_rcnn_resnet101_egohands' and 'faster_rcnn_inception_v2_egohands' models. However, the faster rcnn models are hacky and do not perform well on TX2 yet.
Add support for 'ssd_mobilenet_v2_egohands', 'ssdlite_mobilenet_v2_egohands', 'ssd_inception_v2_egohands', 'faster_rcnn_resnet50_egohands', 'faster_rcnn_resnet101_egohands' and 'faster_rcnn_inception_v2_egohands' models
…TRT) onto GPU, so well as revert number of RPN proposals back to 300; add code to measure tf_sess.run() time
…ection API library: 'detection_boxes', 'detection_scores', 'detection_classes', and 'num_detections'
Add support for 'rfcn_resnet101_egohands', plus some other re-factoring
|
I put my latest code in my own GitHub repository: https://github.com/jkjung-avt/tf_trt_models. Feel free to check it out. Meanwhile, I'm not completely sure what your question is. Are you trying to say that TF-TRT fails to optimize 'faster_rcnn_resnet101' at all? |
…Flow Detection Model Zoo instead)
… snapshot tensorflow 'models'
Update my fork to match the latest code in NVIDIA's original repository
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Hello Mr.Jung,
Tanks to you, I was able to test faster rcnn with tensorrt.
But, the latency is lower than I expected on my machine.
The response time is as follows,
In addition, the memory usage is not much different.
My code is as follows,
trt_graph = trt.create_inference_graph(
input_graph_def=frozen_graph,
outputs=output_names,
max_batch_size=1,
max_workspace_size_bytes=1 << 25,
precision_mode='INT8' #'FP32' / 'FP16',
minimum_segment_size=50
)
Did I do something wrong?
I would really appreciate it if you answer for me.