Differently, our unsupervised multi-body task requires the model's ability to handle part-level local equivariance, Figure 1: Structure of our feature extractor based on EPN. "EPNConv" is the SE(3)-equivariant convolution proposed in the vanilla EPN network. Part-level SE(3)-equivariance is desirable for motion analysis, especially rotation estimation. Song and Y ang utilized the methodology proposed by Choy et al . All other objects were considered part of the background.

Here, we include implementation details omitted from the main paper for brevity. Upon acceptance, a deanonymized repository will be released. The last layer's dimension depended upon the exact The feature extractors and decoders varied by domain. In particular, we found that if we did not apply this linear transformation (i.e., pass the raw encodings For VQ-based methods, use a large enough codebook to have at least one element per class. Other differences simply reflected differences in architecture (e.g., For iNat, we trained all models with batch size 256, using the hyperparameters specified in Table 3.

Under this setting, previous SSL methods tend to predict wrong pseudo-labels with the model fitted on labeled data, resulting in noise accumulation.

We next explain the rationality of the resampling-by-compatibility presented in Sect. We first i) prove Eq. For Assumption (9), L( π, u, v) is strongly convex as proved in [5]. Eq. (A-19), we have E The codes are in pytorch [7]. The learning rate is 1e-5.