Constrain similarity matrix with cameraID.

[windsor718]

In deepreid/identify.py

def match(similarityMatrix, ids, threshold=0.4):
    out = []
    for i in range(similarityMatrix.shape[0]):
        cand_idx = np.where(similarityMatrix[i, :] > threshold)[0]
        print(cand_idx)
        matched = [ids[i]]
        for idx in set(cand_idx) - set([i]):
            print(idx)
            print(similarityMatrix[:, idx])
            max_idx = np.argsort(similarityMatrix[:, idx])[-2]
            if max_idx == i:
                matched.append(ids[idx])
        out.append(matched)
    return out

May need to be constrained by camera id to improve accuracy.
May not be necessary after model re-training?