Length of dataset mismatch

as per the link “https://www.drivendata.org/competitions/58/disaster-response-roof-type/page/143/” , data count would be 22553. But use that benchmark code to extract data gives 17650. almost 5000 image are not extract or missed.

Make sure you have added unverified dataset from gros islet and castries

In the benchmark code we are loading/prepping/training with only 2 regions : “borde_rural” and “borde_soacha” as an example. You can in similar way add all the regions names and extract the data.