Skip to content

Add torch2.6 support for ms_deform_attn_cuda#94

Merged
rentainhe merged 1 commit intoIDEA-Research:mainfrom
QuantuMope:PR/andrew/torch26-deform
Apr 17, 2025
Merged

Add torch2.6 support for ms_deform_attn_cuda#94
rentainhe merged 1 commit intoIDEA-Research:mainfrom
QuantuMope:PR/andrew/torch26-deform

Conversation

@QuantuMope
Copy link
Contributor

Copy link

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot wasn't able to review any files in this pull request.

Files not reviewed (1)
  • grounding_dino/groundingdino/models/GroundingDINO/csrc/MsDeformAttn/ms_deform_attn_cuda.cu: Language not supported

@rentainhe rentainhe merged commit 8238557 into IDEA-Research:main Apr 17, 2025
shojint pushed a commit to shojint/Grounded-SAM-2 that referenced this pull request Sep 2, 2025
shojint pushed a commit to shojint/Grounded-SAM-2 that referenced this pull request Sep 2, 2025
maciejmajek pushed a commit to RobotecAI/Grounded-SAM-2 that referenced this pull request Dec 12, 2025
* support custom video tracking demo with local gd1.0 model

* update dockerfile (IDEA-Research#47)

* feat: grounded_sam2_hf_mode cli arguments (IDEA-Research#52)

* [New Feature] Support SAM 2.1 (IDEA-Research#59)

* support sam 2.1

* refine config path and ckpt path

* update README

* refine config

* fix hyper link for florence-2 demo

* update README

* update running info

* support slice inference on gd1.5 sam2 demo

* update README for SAHI inference

* refine README

* refine README

* refine README

* refine README

* refine useless code

* fix: zero object detection error (IDEA-Research#64)

* update dockerfile

* fix: zero object detection error

* fix: zero object detection error

* support box threshold in GD 1.5 demos

* Fixed Type on MORE_DETAILED_CAPTION (IDEA-Research#68)

* Add DINO-X arXiv link

* support DINO-X with SAM 2 for detection and segmentation

* add dino-x sam2 tracking demo

* Fix typo

* update dds-cloudapi-sdk to 0.3.3

* update README

* update setuptools build requirement to fix build error (IDEA-Research#91)

* Add torch2.6 support for ms_deform_attn_cuda (IDEA-Research#94)

* update DINO-X api usage to dds v2

* update DINO-X api usage to dds v2

* update DINO-X api to V2

* Pinned setuptools in Dockerfile (IDEA-Research#99)

Co-authored-by: Sami Haidar Wehbe <sami@autoenhance.ai>

* feat:add grounded_sam2_tracking_camera_with_continuous_id.py (closes … (IDEA-Research#97)

* feat:add grounded_sam2_tracking_camera_with_continuous_id.py (closes IDEA-Research#74)

* update README

* Change default output dir for HF demo (IDEA-Research#105)

* Fix demos for CPU inference (IDEA-Research#104)

* Fix unsorted `all_class_names` mismatch (IDEA-Research#112)

`all_class_names` was not sorted together along with `all_object_ids` and `all_object_boxes`. This causes issue when drawing boxes on multi-class grounding detection.

* support 5090 (IDEA-Research#115)

* Handle (N,K,H,W) masks when multimask_output=True by picking the top-scoring candidate (IDEA-Research#118)

* Update setup.py

support 5090

* Update grounded_sam2_local_demo.py

fix: handle (N,K,H,W) masks when multimask_output=True by picking the top-scoring candidate

When multimask_output=True, the predictor returns masks with shape (N,K,H,W). The previous code called squeeze(1) assuming K==1 and crashed (“cannot select an axis to squeeze out…”).
This patch selects the best candidate per instance (argmax over scores) to reduce it to (N,H,W).

Note: similar assumptions may exist in other files, but I haven't tested them, so this change is scoped to this file only.

* Replaced 'box_threshold' argument with 'threshold' to processor.post_process_grounded_object_detection for compatibility with the latest version of transformers (IDEA-Research#119)

* fix: allow compilation with lower cuda version (IDEA-Research#122)

* fix: add CUDA version check for <12.8 compatibility (IDEA-Research#123)

Co-authored-by: jieun-b <jieun-b>

---------

Co-authored-by: rentainhe <596106517@qq.com>
Co-authored-by: Susan Shen <31400000+ShuoShenDe@users.noreply.github.com>
Co-authored-by: kwikwag <sadan.yuval@gmail.com>
Co-authored-by: Ren Tianhe <48727989+rentainhe@users.noreply.github.com>
Co-authored-by: John Heilman <1735575+Infro@users.noreply.github.com>
Co-authored-by: Reuben Feinman <rfeinman16@gmail.com>
Co-authored-by: Andrew Choi <neffneff4@gmail.com>
Co-authored-by: Sami Haidar <47423350+sami-haidar@users.noreply.github.com>
Co-authored-by: Sami Haidar Wehbe <sami@autoenhance.ai>
Co-authored-by: Embodied Learner <157788460+HITSZ-Robotics@users.noreply.github.com>
Co-authored-by: will ye <williamyeny@gmail.com>
Co-authored-by: Meng-Jiun Chiou <coldmanck@gmail.com>
Co-authored-by: Freax Ruby <limyoonaxi@gmail.com>
Co-authored-by: Ooosn <namewangyy@gmail.com>
Co-authored-by: Hamza Adnan <42165451+hamzaadnan@users.noreply.github.com>
Co-authored-by: Zicheng Zhang <38455236+victkk@users.noreply.github.com>
Co-authored-by: Jieun Bae <59651199+jieun-b@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants