In few-shot classification, the aim is to learn models able to discriminate\nclasses using only a small number of labeled examples. In this context, works\nhave proposed to introduce Graph Neural Networks (GNNs) aiming at exploiting\nthe information contained in other samples treated concurrently, what is\ncommonly referred to as the transductive setting in the literature. These GNNs\nare trained all together with a backbone feature extractor. In this paper, we\npropose a new method that relies on graphs only to interpolate feature vectors\ninstead, resulting in a transductive learning setting with no additional\nparameters to train. Our proposed method thus exploits two levels of\ninformation: a) transfer features obtained on generic datasets, b) transductive\ninformation obtained from other samples to be classified. Using standard\nfew-shot vision classification datasets, we demonstrate its ability to bring\nsignificant gains compared to other works.\n