UniFormer: Unified Transformer for Efficient Spatiotemporal Representation Learning

Abstract

It is a challenging task to learn rich and multi-scale spatiotemporal semantics from high-dimensional videos, due to large local redundancy and complex global dependency between video frames. The recent advances in this research have been mainly driven by 3D convolutional neural networks and vision transformers. Although 3D convolution can efficiently aggregate local context to suppress local redundancy from a small 3D neighborhood, it lacks the capability to capture global dependency because of the limited receptive field. Alternatively, vision transformers can effectively capture long-range dependency by self-attention mechanism, while having the limitation on reducing local redundancy with blind similarity comparison among all the tokens in each layer. Based on these observations, we propose a novel Unified transFormer (UniFormer) which seamlessly integrates merits of 3D convolution and spatiotemporal self-attention in a concise transformer format, and achieves a preferable balance between computation and accuracy. Different from traditional transformers, our relation aggregator can tackle both spatiotemporal redundancy and dependency, by learning local and global token affinity respectively in shallow and deep layers. We conduct extensive experiments on the popular video benchmarks, e.g., Kinetics-400, Kinetics-600, and Something-Something V1&V2. With only ImageNet-1K pretraining, our UniFormer achieves 82.9%/84.8% top-1 accuracy on Kinetics-400/Kinetics-600, while requiring 10× fewer GFLOPs than other state-of-the-art methods. For Something-Something V1 and V2, our UniFormer achieves new state-of-the-art performances of 60.9% and 71.2% top-1 accuracy respectively. Code is available at https://github.com/Sense-X/UniFormer.

Inception Transformer

“…For image classification, we evaluate iFormer on the ImageNet dataset [28]. We train the iFormer model with the standard procedure in [6,22,29]. Specifically, we use AdamW optimizer with an initial learning rate 1 × 10 −3 via cosine decay [69], a momentum of 0.9, and a weight decay of 0.05.…”

“…Compared with various Transformer backbones, our iFormers still maintain the performance superiority over their results. For example, our iFormer-B surpasses UniFormer-B [22], Swin-S [5] by 0.9 points of AP b and 3.5 points of AP b respectively.…”

“…LV-ViT [50], LeViT [54] and ViT C [21] further stack several layers of convolutions as the stem for models, which is found helpful in training and achieving better performance. Besides the stem, ViT-hybrid [18], CoAtNet [24], Hybrid-MS [55] and UniFormer [22] design early stages with convolution layers. However, the combination of convolution and attention in a serial order means each layer can only process either high or low frequency and neglects the other part.…”

“…Unlike ViTs, they cover more local information through local convolution within the receptive fields, thus effectively extracting high-frequency representations [19,20]. Recent studies [21][22][23][24][25] have integrated CNNs and ViTs considering their complementary advantages. Some methods [21,22,24,25] stack convolution and attention layers in a serial manner to inject the local information into global context.…”

“…Recent studies [21][22][23][24][25] have integrated CNNs and ViTs considering their complementary advantages. Some methods [21,22,24,25] stack convolution and attention layers in a serial manner to inject the local information into global context. Unfortunately, this serial manner only models one type of dependency, either global or local, in one layer, and discards the global information during locality modeling, or vice versa.…”