Résumé
After a brief sketch of covariant loop quantization I'll demonstrate that the consistent implementation of second class constraints in the discretized path integral for 4d general relativity leads to a spin foam model with boundary states identical to the kinematical states of the loop approach. The identification works perfectly well for any signature and any value of the Immirzi parameter. An expression for the vertex amplitude is given in terms of integrated (over a still unknown measure) projected spin networks.