[WIP]: Add DINO on 3.x #8820
[WIP]: Add DINO on 3.x #8820
Conversation
* Update * Update * force reinstall pycocotools * Fix build_cuda * docker install git * Update * comment other job to speedup process * update * uncomment * Update * Update * Add comments for --force-reinstall
…b#8625) * Refactor anchor head * Update * Update * Update * Add a series of boxes tools * Fix box type to support n x box_dim boxes * revert box type changes * Add docstring * refactor retina_head * Update * Update * Fix comments * modify docstring of coder and ioucalculator * Replace with_boxlist with use_box_type
| spatial_shapes = [] | ||
| for lvl, (feat, mask, pos_embed) in enumerate( | ||
| zip(mlvl_feats, mlvl_masks, mlvl_pos_embeds)): | ||
| bs, c, h, w = feat.shape |
There was a problem hiding this comment.
remember to replace all bs with batch_size
|
Hey, what's progress of this PR? |
@austinmw Hi!~ Thanks for your attention! For DINO implementation, our procedure includes three steps:
Besides, there are also brother PRs #8544, #8846, #8533, #8811 of two algorithms (Condition DETR and DAB DETR) developed in parallel. I'm a student contributor in GitLink Code Camp. And this is my first experience of contributing a PR to the open source community. I'm trying hard to make DINO available in mmdetection as soon as possible. If your project allows to use mmdetection2.x, you can refer to #8362 for the version that has been finished but not released. After the official release of DINO in mmdet3.x, you are more advanced to use the new implementation. In the new implementation, the code structure of DETR-like algorithm will be more reasonable and the code readability is improved. Look forward to your attention! |
jshilong
force-pushed
the
refactor-detr
branch
from
af4604f to
cfe9622
Compare
Motivation
Implement the results in the paper "DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detection" with mmdetection. The original code has released.
Remark
This is my first time to submit a PR to reproduces an algorithm, and I'm still lack of experience. I hope that developers who pay attention to this PR can comment more problems in my implementation and share some results. I will also follow up and share some of my own experimental results and progress arrangements.
I have created a PR #8362, which implemented DINO on dev-2.x. This PR is actually a refactor of the previous DINO implementation for dev-3.x and the praevia refactor PR #8763.