Algorithm 3: Co-training Procedure Uses: Algorithms 1 & 2.

Input   : Set of labeled images: ${\mathcal{X}}^l$      Set of unlabeled images: ${\mathcal{X}}^u$      Net. init. weights & training hyp.-p.: $\mathcal{W},{\mathcal{H}}_{\mathcal{W}}$      Self-t. hyp.-p.: ${\mathcal{H}}_{st}=\{\mathcal{T},N,n,K_m,K_M,{\mathcal{M}}_{df}\}$      Co-t. hyp.-p.: ${\mathcal{H}}_{ct}=\{K,w,\lambda \}$ Output: Refined model: $\mathcal{W}$                  // Initialization               ${\mathcal{W}}_{0,1},{\mathcal{W}}_{0,2}$ ←$\mathrm{SelfTraining}\left({\mathcal{X}}^l,{\mathcal{X}}^u,\mathcal{W},{\mathcal{H}}_{\mathcal{W}},{\mathcal{H}}_{st}\right)$ ${\mathcal{X}}_1^{\widehat{l}},{\mathcal{X}}_2^{\widehat{l}},k,{{\mathcal{V}}_{C_T}}_1,{{\mathcal{V}}_{C_T}}_2,{\mathcal{W}}_1,{\mathcal{W}}_2$ ←$\varnothing ,\varnothing ,0,\mathbf{0},\mathbf{0},{\mathcal{W}}_{0,1},{\mathcal{W}}_{0,2}$

$\mathcal{W}$←$\mathrm{LastTrain}(w,{\mathcal{W}}_1,{\mathcal{W}}_2,{\mathcal{H}}_{\mathcal{W}},{\mathcal{X}}^l,{\mathcal{X}}^u,{\mathcal{M}}_{df})$ return$\mathcal{W}$