Monte Carlo based methods such as path tracing are widely used in movie production. To achieve noise-free quality, they tend to require a large number of samples per pixel, resulting in longer rendering time. To reduce that cost, one solution is Monte Carlo denoising: render the image with fewer samples per pixel (as little as 128) and then denoise the resulting image. Many Monte Carlo denoising methods rely on deep learning: they use convolutional neural networks to learn the relationship between noisy images and reference images, using auxiliary features such as position and normal together with image color as inputs. The network predicts kernels which are then applied to the noisy input. These methods have shown powerful denoising ability. However, they tend to lose geometric details or lighting details and over blur sharp features during denoising. In this paper, we solve this issue by proposing a novel network structure, a new input feature-light transport covariance from path space-and an improved loss function. In our network, we separate feature buffers with color buffer to enhance detail effects. Their features are extracted separately and then are integrated to a shallow kernel predictor. Our loss function considers perceptual loss, which also improves the detail preserving. In addition, we present the light transport covariance feature in path space as one of the features, which is used to preserve illumination details. Our method denoises Monte Carlo path traced images while preserving details much better than previous work.