B-Spline CNNs on Lie groups
Authors: Erik J Bekkers
ICLR 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The impact and potential of our approach is studied on two benchmark datasets: cancer detection in histopathology slides in which rotation equivariance plays a key role and facial landmark localization in which scale equivariance is important. In both cases, GCNN architectures outperform their classical 2D counterparts and the added value of atrous and localized group convolutions is studied in detail. |
| Researcher Affiliation | Academia | Erik J. Bekkers Amsterdam Machine Learning Lab Informatics Institute University of Amsterdam e.j.bekkers@uva.nl Centre for Analysis and Scientific Computing Applied Mathematics and Computer Science Eindhoven University of Technology |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Implementations are available at https://github.com/ebekkers/gsplinets. |
| Open Datasets | Yes | The Patch Camelyon (PCam) dataset (Veeling et al., 2018) consists of 327,680 RGB patches taken from histopathologic scans of lymph node sections and is derived from Camelyon16 (Ehteshami Bejnordi et al., 2017). The Celeb A dataset (Liu et al., 2015) contains 202,599 RGB images of varying size of celebrities together with labels for attributes (hair color, glasses, hat, etc) and 5 annotated facial landmarks... |
| Dataset Splits | No | The paper describes experimental setups, including network architectures and data processing, but does not explicitly provide specific train/validation/test dataset splits (e.g., percentages or counts). |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU/GPU models, memory specifications) used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific software dependency details (e.g., library or solver names with version numbers). |
| Experiment Setup | Yes | The architecture for metastasis classification in the PCam dataset is given in Tab. 1. The input (64 64) is first cropped to 88 88 and is then used as input for the first layer (the lifting layer). None of the layers use spatial padding such that the image is eventually cropped to size 1 1. Each layer is followed by batch normalization and a ReLU activation function, except for the last layer (layer 7) which is followed by adding a bias vector of length 2 and a softmax. ... The architecture for landmark detecion in the Celeb A dataset is biven in Tab. 2. The input is formatted according to the details in Sec. 4. In each layer zero padding is used in order to map the 128 128 input images to a 128 128 output heatmaps. Each layer is followed by batch normalization and a Re LU activation function, except for the last layer (layer 10) which is followed by adding a bias vector and a logistic sigmoid activation function. |