4.2.2. Training and Inference Details The experimental environment is a setup with an Intel Core i9-10900K [email protected] GHz, NVIDIA GeForce RTX 3090, 64 GB of RAM, and the operating system is Ubuntu 18.04. The model uses ResNet-50 [31] pre-trained by ImageNet [32] and a five-layer feature pyramid structure as the backbone network, both with a training period of 1×. According to the linear strategy, the batch size of training was adjusted to eight for 180 k iterations (i.e., 12 epochs).
