Support Inference over batch with TensorRT Engine Model

Use dynamic batch size by default

Add script used to convert ONNX to fp32/fp16/int8/mixed engine

Use ONNXruntime instead of ONNX.checker.check_model to detect ONNX model

Resize in keeping aspect ratio in visualization

fix postprocess_batch method in RTMO_GPU_Batch class

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

Add support TensorRT engine support for RTMO

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

Improve Debug Message Layout in Validate Funciton

Add appropriate pre- to input image and postprocessing to validate function

Validate only on pose result, add visual check

Add comparison between joint coordinated from original and converted models

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

Fix Input Type Error raised in infer()

Add utility to convert ONNX model in FP32/16 mixed precision

Load ./libmmdeploy_tensorrt_ops.so for TensorRT EP if available

Remove fix_batch_dimension_all.sh

Remove models with fixed batch size as dynamic batch is generally supported

Apply constant folding

Reperform shape inference on all models

Bugfix: correct batch dimension fixation then regenerate ONNX models

Update README.md

Update README.md

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

Fix typo in model names mentioned in demo.sh demo_batch.sh

Repair DEMO to run with TRT EP

Fix FP16 models having poor accuracy

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

Make ONNX models compatible with ONNXruntime's TensorrtExecutionProvider

Use RTMO_GPU_Batch to demo

Add new class 'RTMO_GPU_Batch' that can perform inference on batch of images

Optimization for CUDAExecutionProvider

INT8 models obtained via dynamic quantization

Update README.md

Add a sample video

RTMO ONNX Models from MMPOSE range from Tiny to Large

initial commit