Commit History

  1. Increase logging level to extra verbose

    a9584a8

  1. Increase logging level to verbose on onnx to engine script

    3deadf3
  2. trim sample video to keep only important part

    luigi
    f5e3a24
  3. Make 2 distinct ONNX2TRT conversion scripts, one for JetPack 4.6, the other for JetPack 5.1

    06f3435

  1. adjust visualisation style, allow draw_bbox() to show person id

    a9e21bc

  1. Add method to free buffers

    b79a8af

  1. Implement pycuda backend for inference with TensorRT engine⁠

    889281f

  1. Add 'get_model_format_and_input_shape' helper function

    Luigi
    2f2f685
  2. Return and show bounding box confidence

    634d4ff

  1. Deal with multiple cameras

    Luigi
    98d66d4

  1. Show bounding box on screen too

    Luigi
    0bf1eb7
  2. Internalize input frame buffer

    Luigi
    846e714
  3. Unify batched and non-batched versions

    Luigi
    42892de
  4. Show FPS on demo video

    15801f5
  5. Support Inference over batch with TensorRT Engine Model

    0cdc9a7
  6. Use dynamic batch size by default

    09ccc6e
  7. Add script used to convert ONNX to fp32/fp16/int8/mixed engine

    92676db
  8. Use ONNXruntime instead of ONNX.checker.check_model to detect ONNX model⁠

    57a8c6b

  1. Resize in keeping aspect ratio in visualization

    35b2a45
  2. fix postprocess_batch method in RTMO_GPU_Batch class

    6008f96

  1. Add model format (ONNX/Engine) & input size detection (based on file header, not on filename) for RTMO

    58a44cf

  1. Add support TensorRT engine support for RTMO

    bbf20b6

  1. Re-quantize models in FP16 in keep positional encoding in FP32 to avoid bad accuracy

    Verified signature
    f9a6075

  1. Improve Debug Message Layout in Validate Funciton

    e8f08e3
  2. Add appropriate pre- to input image and postprocessing to validate function

    6dfe441

  1. Validate only on pose result, add visual check

    b58a63b
  2. Add comparison between joint coordinated from original and converted models

    ab55aa6
  3. Add 'rtol', 'atol' arguments to control tolerance in model validation

    684ebde
  4. Fix Input Type Error raised in infer()

    18eb8e3
  5. Add utility to convert ONNX model in FP32/16 mixed precision

    9f60e86

  1. Load ./libmmdeploy_tensorrt_ops.so for TensorRT EP if available

    1c5dc58
  2. Remove fix_batch_dimension_all.sh

    01042ef
  3. Remove models with fixed batch size as dynamic batch is generally supported

    ae1686f

  1. Apply constant folding

    d474b05
  2. Reperform shape inference on all models

    e7b68a5

  1. Bugfix: correct batch dimension fixation then regenerate ONNX models⁠

    Luigi
    e4e03fd

  1. Update README.md

    Verified signature
    bd45a81
  2. Update README.md

    Verified signature
    0a36dd1

  1. Add option --yolo_nas_pose, used to read YOLO NAS Pose model instead of RTMO

    1f0f5d8
  2. Fix typo in model names mentioned in demo.sh demo_batch.sh

    9133777
  3. Repair DEMO to run with TRT EP

    c4f84a9
  4. Fix FP16 models having poor accuracy⁠

    25f1444

  1. Bugfix for batch demo: add padding if remaning number of frames is not equal to batch size

    12e168d
  2. Make ONNX models compatible with ONNXruntime's TensorrtExecutionProvider⁠

    b75f05d

  1. Use RTMO_GPU_Batch to demo

    3939a91
  2. Add new class 'RTMO_GPU_Batch' that can perform inference on batch of images

    8072759
  3. Optimization for CUDAExecutionProvider⁠

    43a489d

  1. INT8 models obtained via dynamic quantization

    Verified signature
    e1ce29b

  1. Update README.md

    Verified signature
    c4cff5e
  2. Add a sample video

    Verified signature
    2e05515