Improving Latent Quantization of Learned Image Compression with Gradient Scaling | IEEE Conference Publication | IEEE Xplore