This is a CNN+RNN based model that predicts the position of the object in sequence
The model consists of a pre-trained ResNet18 from torchvision, and an LSTM model. The last two layers of the ResNet18 is deleted and replaced with a Adaptive Average Pooling layer which outputs a tensor with dimension [B*T 512 2 2]. Then the tensor is flattened and sent to a Linear layer followed by a ReLU and Dropout layer for down sampling. Then the LSTM model with hidden_size 256 and num_layers=2 followed by a Linear Layer outputs the prediction. The prediction is a tensor with four elements `[delta x, delta y, delta w, delta h]
