Training Image Estimators without Image Ground Truth
Authors: Zhihao Xia, Ayan Chakrabarti
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our method for training networks for compressive-sensing and blind deconvolution, considering both non-blind and blind training for the latter. Our unsupervised framework yields models that are nearly as accurate as those from fully supervised training, despite not having access to any ground-truth images. We validate our method with experiments on image reconstruction from compressive measurements and on blind deblurring of face images, with blind and non-blind training for the latter, and compare to fully-supervised baselines with state-of-the-art performance. |
| Researcher Affiliation | Academia | Zhihao Xia Washington University in St. Louis 1 Brookings Dr., St. Louis, MO 63130 zhihao.xia@wustl.edu Ayan Chakrabarti Washington University in St. Louis 1 Brookings Dr., St. Louis, MO 63130 ayan@wustl.edu |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The source code of our implementation is available at https://projects.ayanc.org/unsupimg/. |
| Open Datasets | Yes | We generate a training and validation set, of 100k and 256 images respectively, by taking 363 × 363 crops from images in the Image Net database [26]. We use all 160k images in the Celeb A training set [17] and 1.8k images from Helen training set [13] to construct our training set, and 2k images from Celeb A val and 200 from the Helen training set for our validation set. |
| Dataset Splits | Yes | We generate a training and validation set, of 100k and 256 images respectively, by taking 363 × 363 crops from images in the Image Net database [26]. We use all 160k images in the Celeb A training set [17] and 1.8k images from Helen training set [13] to construct our training set, and 2k images from Celeb A val and 200 from the Helen training set for our validation set. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU or CPU models used for running the experiments. |
| Software Dependencies | No | The paper does not explicitly list specific software dependencies with version numbers (e.g., Python 3.x, PyTorch 1.x). |
| Experiment Setup | Yes | The weight γ for the self-measurement loss is set to 0.05 based on the validation set. The weights α, β, γ are all set to one in this case. We use a CNN architecture that stacks two UNets [24], with a residual connection between the two (see supplementary). Then, for unsupervised training with our approach, we choose two kernels for each training image to form a training set of measurement pairs, that are kept fixed (including the added Gaussian noise) across all epochs of training. |