Please provide the following information when requesting support.
• Hardware (x86_64)
• Network Type (RTDetr)
Good day. I decided to run a model downloaded from the NGC catalog ( GPU-optimized AI, Machine Learning, & HPC Software | NVIDIA NGC | NVIDIA NGC) in tao deploy. I converted it using tao gen_trt_engine and ran it through tao deploy, but instead of annotated images, it outputs a whole bunch of incorrect images that are not related to the data used.
gen_trt_yaml
gen_trt_engine:
results_dir: null
gpu_id: 0
onnx_file: ???
trt_engine: ???
timing_cache: null
batch_size: -1
verbose: false
tensorrt:
workspace_size: 1024
min_batch_size: 1
opt_batch_size: 1
max_batch_size: 4
layers_precision: []
data_type: FP32
calibration:
cal_image_dir: ???
cal_cache_file: ???
cal_batch_size: 1
cal_batches: 1
command:
!tao deploy rtdetr gen_trt_engine -e $SPECS_DIR/gen_trt_engine.yaml \
results_dir=$RESULTS_DIR/gen_trt_engine \
gen_trt_engine.onnx_file=$RESULTS_DIR/export/rtdetr_warehouse_v1.0.onnx \
gen_trt_engine.trt_engine=$RESULTS_DIR/gen_trt_engine/rtdetr_warehouse.engine \
gen_trt_engine.tensorrt.data_type=FP16 \
gen_trt_engine.results_dir=$RESULTS_DIR/gen_trt_engine \
gen_trt_engine.tensorrt.calibration.cal_image_dir=[warehouse_rtdetr/forklift_person/test] \
gen_trt_engine.tensorrt.calibration.cal_cache_file=$RESULTS_DIR/gen_trt_engine
infer.yaml
inference:
conf_threshold: 0.5
results_dir: ???
input_width: 960
input_height: 544
trt_engine: ???
color_map:
person: red
forklift: green
dataset:
infer_data_sources:
image_dir: warehouse_rtdetr/forklifts/
classmap: ???
num_classes: 2
batch_size: 1
command:
!tao deploy rtdetr inference -e $SPECS_DIR/infer.yaml \
inference.trt_engine=$RESULTS_DIR/gen_trt_engine/rtdetr_warehouse.engine \
inference.results_dir=$RESULTS_DIR/gen_trt_engine/result \
dataset.infer_data_sources.classmap=$SPECS_DIR/classmap.txt
Result:
But this is not the data that I send to the model