Region-Aware Pretraining for Open-Vocabulary Object Detection with Vision Transformers
This addresses the challenge of adapting vision transformers for open-vocabulary detection, offering incremental improvements over existing methods.
The paper tackles the problem of bridging image-level pretraining with open-vocabulary object detection by proposing a region-aware pretraining method, achieving state-of-the-art results of 34.1 AP_r on LVIS and improvements in image-text retrieval.
We present Region-aware Open-vocabulary Vision Transformers (RO-ViT) - a contrastive image-text pretraining recipe to bridge the gap between image-level pretraining and open-vocabulary object detection. At the pretraining phase, we propose to randomly crop and resize regions of positional embeddings instead of using the whole image positional embeddings. This better matches the use of positional embeddings at region-level in the detection finetuning phase. In addition, we replace the common softmax cross entropy loss in contrastive learning with focal loss to better learn the informative yet difficult examples. Finally, we leverage recent advances in novel object proposals to improve open-vocabulary detection finetuning. We evaluate our full model on the LVIS and COCO open-vocabulary detection benchmarks and zero-shot transfer. RO-ViT achieves a state-of-the-art 34.1 $AP_r$ on LVIS, surpassing the best existing approach by +7.8 points in addition to competitive zero-shot transfer detection. Surprisingly, RO-ViT improves the image-level representation as well and achieves the state of the art on 9 out of 12 metrics on COCO and Flickr image-text retrieval benchmarks, outperforming competitive approaches with larger models.