Fig. 1. The imaging model of SFS^[35]

Fig. 2. The imaging model of photometric stereo

Fig. 3. The illustration of photometric stereo 3D reconstruction^[43]. (a)-(e) The synthetic images I1(x,y)-I5(x,y) of a sphere; (f) the reconstructed albedo ρ(x,y); (g) the reconstructed vector field g(x,y); (h) the reconstructed normal field N(x,y); (i) the reconstructed shape z(x,y) of the sphere

Fig. 4. The illustration of the commonly used synthetic photometric stereo datasets. (a) Blobby photometric stereo dataset^[72];(b) Sculpture photometric stereo dataset^[72]; (c) CyclesPS photometric stereo dataset^[79]

Fig. 5. Schematic diagrams of DiLiGenT dataset^[62-63]. (a) Photometric stereo images under a certain light direction; (b) surface normal of the objects

Fig. 6. The architecture of DPSN^[69]

Fig. 7. Reconstruction results for several objects from DiLiGenT dataset^[69]

Fig. 8. The architecture of PS-FCN^[72]

Fig. 9. Visualization of the feature map after fusion^[72]. (a)-(f) 6 of 128 channels of the fused feature map

Fig. 10. The architecture of SDPS-Net^[73]

Fig. 11. The coordinate system for observation map

Fig. 12. The framework of the prediction module of CNN-PS^[79]

Fig. 13. The architecture of SPLINE-Net^[80]

Fig. 14. The architecture of LMPS^[75]

Fig. 15. The architecture of DR-PSN^[77]

Fig. 16. The architecture of IRPS^[114]