We provide a baseline implementation. It is a improved version based on the previous baseline[tag]. It can reach 0.5 top1ratio after 2 days' training and 0.8 in 4 days, using 1 V100 and 8 cpu cores. The following is the training curve
We provide a baseline implementation. It is a improved version based on the previous baseline[tag]. It can reach 0.8 top1ratio in 2 days' training, using 1 V100 and 8 cpu cores. The following is the training curve