Skip to content

Optimize RF-DETR Triton preprocessing #45

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
aseembits93 wants to merge 12 commits into
base: opt-python-postproc
Choose a base branch
from
Open

Optimize RF-DETR Triton preprocessing #45

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
0558da5
perf(rfdetr-seg): fused Triton preproc kernel for TRT path
aseembits93 May 9, 2026
1b8d6c1
perf(rfdetr): add kill switch for Triton preproc fast path
aseembits93 May 11, 2026
f959d81
Optimize RF-DETR Triton preprocessing
aseembits93 May 29, 2026
be8e0ff
Consolidate RF-DETR Triton preproc kernel
aseembits93 Jun 2, 2026
488a703
Align Triton preproc env integration
aseembits93 Jun 2, 2026
6441d8e
Tighten RF-DETR Triton preprocessing runtime
aseembits93 Jun 2, 2026
c8d59a7
Tighten RF-DETR Triton preproc review coverage
aseembits93 Jun 3, 2026
a923560
Add RF-DETR Triton preproc integration coverage
aseembits93 Jun 3, 2026
6bfbbaa
Clarify RF-DETR Triton preproc runtime
aseembits93 Jun 3, 2026
ab811f2
Tighten RF-DETR Triton preproc review coverage
aseembits93 Jun 3, 2026
85689ec
Gate RF-DETR preprocess by preproc flag
aseembits93 Jun 4, 2026
18da99e
Revert "Gate RF-DETR preprocess by preproc flag"
aseembits93 Jun 4, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 5 additions & 4 deletions development/stream_interface/rfdetr_coco_same_shape_parity.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,16 +1,17 @@
"""Compare RF-DETR instance-segmentation outputs on same-shape COCO images.

This harness is used to reproduce the correctness table in the RF-DETR Triton
postprocess PR. It runs a baseline git ref with all RF-DETR fast paths disabled
and a candidate ref with only Triton RLE postprocess enabled, then compares
preprocess PR. It runs a baseline git ref with all RF-DETR fast paths disabled
and a candidate ref with Triton RLE postprocess and Triton preprocess enabled,
then compares
detection counts, classes, boxes, scores, and RLE masks.

Example:

env PARITY_MODEL_PATH=/path/to/rfdetr-seg-nano-orin-trt-package \
python development/stream_interface/rfdetr_coco_same_shape_parity.py \
--base-ref main \
--candidate-ref opt-python-postproc \
--candidate-ref opt-preprocess \
--height 480 \
--width 640 \
--image-count 1000
Expand Down Expand Up @@ -47,7 +48,7 @@
}
CANDIDATE_FLAGS_ON = {
"INFERENCE_MODELS_RFDETR_TRITON_POSTPROC_ENABLED": "true",
"INFERENCE_MODELS_RFDETR_TRITON_PREPROC_ENABLED": "false",
"INFERENCE_MODELS_RFDETR_TRITON_PREPROC_ENABLED": "true",
"RFDETR_PIPELINE_DEPTH": "1",
"ENABLE_AUTO_CUDA_GRAPHS_FOR_TRT_BACKEND": "false",
"RFDETR_NSIGHT_MARKERS": "false",
Expand Down
Loading