Unsupervised learning of object frames by dense equivariant image labelling
Authors: James Thewlis, Hakan Bilen, Andrea Vedaldi
NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the applicability of this method to simple articulated objects and deformable objects such as human faces, learning embeddings from random synthetic transformations or optical flow correspondences, all without any manual supervision. This section assesses our unsupervised method for dense object labelling on two representative tasks: two toy problems (sections 4.1 and 4.2) and human and cat faces (section 4.3). |
| Researcher Affiliation | Academia | 1 Visual Geometry Group University of Oxford {jdt,vedaldi}@robots.ox.ac.uk, 2 School of Informatics University of Edinburgh hbilen@ed.ac.uk |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described, nor does it explicitly state that the code is available. |
| Open Datasets | Yes | We train our models on the extensive Celeb A [26] dataset of over 200k faces as in [39], excluding MAFL [49] test overlap from the given training split. It has annotations of the eyes, nose and mouth corners. Note that we do not use these to train our model. We also use AFLW [23], testing on 2995 faces [49, 42, 48] with 5 landmarks. |
| Dataset Splits | No | The paper mentions a 'training subset' (MAFL training subset of 19k images) and a 'test split' (standard test split for evaluation of the MAFL dataset [49], containing 1000 images) but does not provide explicit details for a validation split or combined train/validation/test percentages/counts that would enable full reproduction of the data partitioning for all three phases. |
| Hardware Specification | No | The paper does not provide specific details regarding the hardware (e.g., GPU/CPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions 'Cap Sim the MATLAB physics engine' in a reference, implying MATLAB, but does not specify a version number for MATLAB or any other software dependencies crucial for replication. |
| Experiment Setup | Yes | Here γ > 0 is an exponent used to control the robustness of the distance measure, which we set to γ = 0.5, 1. Network details. We test two architecture. The first one, denoted SIMPLE, is the same as [49, 39] and is a chain (5, 20)+, (2, mp), 2, (5, 48)+, (3, 64)+, (3, 80)+, (3, 256)+, (1, 3) where (h, c) is a bank of c filters of size h h, + denotes Re LU, (h, mp) is h h max-pooling, s is s downsampling. Better performance can be obtained by increasing the support of the filters in the network; for this, we consider a second network DILATIONS (5, 20)+, (2, mp), 2 , (5, 48)+, (5, 64, 2)+, (3, 80, 4)+, (3, 256, 2)+, (1, 3) where (h, c, d) is a filter with d dilation [43]. |