I have dataset include $960$ train images and $240$ test image with $4$ classes $"$ Action", "Horror", "Romance" and "Comedy". I trained with CNN model, then I used the EfficientNet$\_$B$0$ as the base model, perform transfer learning to re$-$train the final layer using

this new dataset and achieve $67\mathrm{.}92\%$ test accuracy. Now I create $1$ image and want to see if the efficientnet model identifies them correctly or not. How can I do it$?$ Do you need I provide model?