Open barbolo opened 7 years ago
http://science.sciencemag.org/content/sci/suppl/2017/10/25/science.aag2612.DC1/aag2612_George_SM.pdf There are some discussion about big real dataset in section 8.13.
Even though the experiment of Section 8.12 shows that the current incarnation of RCN can also be applied successfully to object detection and instance segmentation, RCN does not produce state-of-the-art results on real-world images such as those from ImageNet. As we have shown in previous sections, RCN derives distinct advantages from being a highly-structured generative model, but additional pieces are needed to make RCN’s detection performance competitive in those benchmarks: Instance merging, Use of appearance during the forward pass, ......
Should I expect to have better than CNN accuracy when using larger datasets?