Roadmap of detrex

[rentainhe]

We keep this issue open to collect feature requests from users and hear your voice.

You can either:

• Suggest a new feature by leaving a comment.
• Vote for a feature request with 👍. (Remember that developers are sometimes busy and cannot respond to all feature requests, so vote for your most favorable one!)
• Tell us that you would like to help implement new features or review the PRs. (This is the greatest thing to hear about!)

v0.2.1

• MaskDINO. (Pre-release in Support MaskDINO COCO instance/panoptic segmentation #154)
• Support SwinV2 backbone ([Feature] Support swin-v2 backbone #152)
• Add a basic usage notebook for inference on custom images. ( [Features] Add demo for visualizing on customized image or video #119 )
• ViT backbone ([Feature] Support ViT backbone and release DINO new baselines #138)

Collection

• DQ-DET: https://arxiv.org/pdf/2307.12239.pdf
• RT-DETR
• Co-DETR
• Better data augmentation code format: [Feature] Add AutoAugment and clean dataset mapper #204
• Support wandb to log training informations and visualize sampled results
• TensorRT for fast inference
• ONNX for DETR models
• Docker or python package for detrex (High Priority)
• Object365 Dataloader
• Semantic Seg/Instance Seg/Key points tasks

[caopulan]

Wandb to log training process and visualize the sampled results.

[ray-lee-94]

support tensorRT for fast inference

[wangxuewen99]

mask-DINO

[simonJJJ]

Objects365 dataloader.

[Hezhexi2002]

support multiple downstream tasks like keypoint detection or segmention and so on

[helq2612]

vote for "Semantic Seg/Instance Seg/Key points tasks"
Also hope this framework can support other backbones, e.g. ViT

[zldrobit]

Would you like to upload pre-trained weights on objects365, so we could reproduce the training for COCO dataset? Or we could fine tune Objects365 pre-trained weights on custum datasets.

[czx94]

We want ONNX!

[wangzhaoyang-508]

hope support salience DETR