I have met a similar error, and I think you can check the multiply operation responding to “gender_resnet0_batchnorm0/mul”.
It is supported in tensorflow if two multipliers with shape [1,32, 3, 2] and [3,2], but it is not supported in tensorrt. Then you can change the shape of the second multiplier to [1, 32, 3, 2], now you have two multiplier with shape[1, 32, 3, 2] and [1, 32, 3, 2], then the multiply operation can be supported in tensorrt, and the result will be the same.