Video-based person re-identification by intra-frame and inter-frame graph neural network

Highlights：

• The intra-frame and inter-frame graph embedding modules treat part features from each frame as graph nodes, and learn video representations by conducting graph convolution.

• This paper proposes a novel occlusion-invariant term to make the part features close to their center, which can relive several uncontrolled complicated factors, such as occlusion and pose invariance.

• We adopts projection metric learning on Grassman manifold to measure the similarities between learned pedestrian features, which has stronger capability in un-linear feature space.

• We carryout extensive experiments on fourwidely used datasetsofMARS, DukeMTMC-VideoReID, PRID2011 and iLID-VID. The experimental results demonstrate that our proposed I2GNN model is more competitive than other state-of-the-art methods.

摘要

•In order to take advantage of the relationship among different parts, this paper proposes a novel Intra-frame and Inter-frame Graph Neural Network (I2GNN) to solve the video-based person Re-ID task.•The intra-frame and inter-frame graph embedding modules treat part features from each frame as graph nodes, and learn video representations by conducting graph convolution.•This paper proposes a novel occlusion-invariant term to make the part features close to their center, which can relive several uncontrolled complicated factors, such as occlusion and pose invariance.•We adopts projection metric learning on Grassman manifold to measure the similarities between learned pedestrian features, which has stronger capability in un-linear feature space.•We carryout extensive experiments on fourwidely used datasetsofMARS, DukeMTMC-VideoReID, PRID2011 and iLID-VID. The experimental results demonstrate that our proposed I2GNN model is more competitive than other state-of-the-art methods.