vectorSearchScore when using $vectorSearch is always around 0.62

I am using $vectorSearch and score: { $meta: ‘vectorSearchScore’ } to get the similarity score, but the score is always around 0.62.

I have filled the collection with image embeddings from the CLIP model to be enable text search on images.

this is the options when creating the index:

{
    name: 'clip-embeddings',
    definition: {
      mappings: {
        fields: {
          embedding: {
            type: 'knnVector',
            dimensions: 512,
            similarity: 'cosine',
          },
          userId: {
            type: 'string',
          },
        },
      },
    },
  }

Anyone have any idea how to debug this, or what can be wrong ? I have around 200 documents in the indexed collection and this is my aggregation query:

[
          {
            $vectorSearch: {
              index: 'clip-embeddings',
              queryVector: queryVector,
              path: 'embedding',
              numCandidates: 100,
              limit: 30,
            },
          },
          {
            $project: {
              _id: 1,
              score: { $meta: 'vectorSearchScore' },
            },
          },
        ]

Results:

[
  {
    _id: new ObjectId("65759305cf078e2685bf784d"),
    score: 0.6358944177627563
  },
  {
    _id: new ObjectId("657592facf078e2685bf76a0"),
    score: 0.6316871643066406
  },
  {
    _id: new ObjectId("657592fccf078e2685bf76c1"),
    score: 0.6295642852783203
  },
  {
    _id: new ObjectId("657592f6cf078e2685bf7640"),
    score: 0.6286431550979614
  },
  {
    _id: new ObjectId("65759307cf078e2685bf788f"),
    score: 0.62856125831604
  },
  {
    _id: new ObjectId("657592fdcf078e2685bf76f3"),
    score: 0.6276072263717651
  },
  {
    _id: new ObjectId("657592facf078e2685bf7691"),
    score: 0.6264308094978333
  },
  {
    _id: new ObjectId("65759300cf078e2685bf776f"),
    score: 0.6250327825546265
  },
  {
    _id: new ObjectId("657592f4cf078e2685bf761f"),
    score: 0.6245126724243164
  },
  {
    _id: new ObjectId("657592f8cf078e2685bf7660"),
    score: 0.6242277026176453
  },
  {
    _id: new ObjectId("65759304cf078e2685bf7822"),
    score: 0.6236228942871094
  },
  {
    _id: new ObjectId("657592fccf078e2685bf76d6"),
    score: 0.6234565377235413
  },
  {
    _id: new ObjectId("65759302cf078e2685bf77ee"),
    score: 0.6234076023101807
  },
  {
    _id: new ObjectId("657592fccf078e2685bf76cb"),
    score: 0.6229453682899475
  },
  {
    _id: new ObjectId("657592ffcf078e2685bf7749"),
    score: 0.6228808760643005
  },
  {
    _id: new ObjectId("657592ffcf078e2685bf7759"),
    score: 0.6227925419807434
  },
  {
    _id: new ObjectId("65759303cf078e2685bf77ff"),
    score: 0.6221488118171692
  },
  {
    _id: new ObjectId("657592f8cf078e2685bf7664"),
    score: 0.6215692758560181
  },
  {
    _id: new ObjectId("65759301cf078e2685bf77d4"),
    score: 0.621394157409668
  },
  {
    _id: new ObjectId("657592f5cf078e2685bf763a"),
    score: 0.6213436722755432
  },
  {
    _id: new ObjectId("65759300cf078e2685bf7783"),
    score: 0.621039628982544
  },
  {
    _id: new ObjectId("65759302cf078e2685bf77e9"),
    score: 0.6207945346832275
  },
  {
    _id: new ObjectId("657592f6cf078e2685bf7646"),
    score: 0.6207586526870728
  },
  {
    _id: new ObjectId("65759300cf078e2685bf7795"),
    score: 0.6205947995185852
  },
  {
    _id: new ObjectId("65759300cf078e2685bf779c"),
    score: 0.6205136775970459
  },
  {
    _id: new ObjectId("657592f2cf078e2685bf75fd"),
    score: 0.6204417943954468
  },
  {
    _id: new ObjectId("657592fccf078e2685bf76d7"),
    score: 0.6201877593994141
  },
  {
    _id: new ObjectId("65759306cf078e2685bf7871"),
    score: 0.6200750470161438
  },
  {
    _id: new ObjectId("65759305cf078e2685bf7846"),
    score: 0.6199190616607666
  },
  {
    _id: new ObjectId("657592fccf078e2685bf76dd"),
    score: 0.6196751594543457
  }
]