[2/2] KNN classifier (lands dataset)

{"id": "itzURpN5wbUNOXOw", "meta": {"instanceId": "205b3bc06c96f2dc835b4f00e1cbf9a937a74eeb3b47c99d0c30b0586dbf85aa"}, "name": "[2/2] KNN classifier (lands dataset)", "tags": [{"id": "QN7etptCmdcGIpkS", "name": "classifier", "createdAt": "2024-12-08T22:08:15.968Z", "updatedAt": "2024-12-09T19:25:04.113Z"}], "nodes": [{"id": "33373ccb-164e-431c-8a9a-d68668fc70be", "name": "Embed image", "type": "n8n-nodes-base.httpRequest", "position": [-140, -240], "parameters": {"url": "https://api.voyageai.com/v1/multimodalembeddings", "method": "POST", "options": {}, "jsonBody": "={{\n{\n \"inputs\": [\n {\n \"content\": [\n {\n \"type\": \"image_url\",\n \"image_url\": $json.imageURL\n }\n ]\n }\n ],\n \"model\": \"voyage-multimodal-3\",\n \"input_type\": \"document\"\n}\n}}", "sendBody": true, "specifyBody": "json", "authentication": "genericCredentialType", "genericAuthType": "httpHeaderAuth"}, "credentials": {"httpHeaderAuth": {"id": "", "name": "[Your httpHeaderAuth]"}}, "typeVersion": 4.2}, {"id": "58adecfa-45c7-4928-b850-053ea6f3b1c5", "name": "Query Qdrant", "type": "n8n-nodes-base.httpRequest", "position": [440, -240], "parameters": {"url": "={{ $json.qdrantCloudURL }}/collections/{{ $json.collectionName }}/points/query", "method": "POST", "options": {}, "jsonBody": "={{\n{\n \"query\": $json.ImageEmbedding,\n \"using\": \"voyage\",\n \"limit\": $json.limitKNN,\n \"with_payload\": true\n}\n}}", "sendBody": true, "specifyBody": "json", "authentication": "predefinedCredentialType", "nodeCredentialType": "qdrantApi"}, "credentials": {"qdrantApi": {"id": "", "name": "[Your qdrantApi]"}}, "typeVersion": 4.2}, {"id": "258026b7-2dda-4165-bfe1-c4163b9caf78", "name": "Majority Vote", "type": "n8n-nodes-base.code", "position": [840, -240], "parameters": {"language": "python", "pythonCode": "from collections import Counter\n\ninput_json = _input.all()[0]\npoints = input_json['json']['result']['points']\nmajority_vote_two_most_common = Counter([point[\"payload\"][\"landscape_name\"] for point in points]).most_common(2)\n\nreturn [{\n \"json\": {\n \"result\": majority_vote_two_most_common \n }\n}]\n"}, "typeVersion": 2}, {"id": "e83e7a0c-cb36-46d0-8908-86ee1bddf638", "name": "Increase limitKNN", "type": "n8n-nodes-base.set", "position": [1240, -240], "parameters": {"options": {}, "assignments": {"assignments": [{"id": "0b5d257b-1b27-48bc-bec2-78649bc844cc", "name": "limitKNN", "type": "number", "value": "={{ $('Propagate loop variables').item.json.limitKNN + 5}}"}, {"id": "afee4bb3-f78b-4355-945d-3776e33337a4", "name": "ImageEmbedding", "type": "array", "value": "={{ $('Qdrant variables + embedding + KNN neigbours').first().json.ImageEmbedding }}"}, {"id": "701ed7ba-d112-4699-a611-c0c134757a6c", "name": "qdrantCloudURL", "type": "string", "value": "={{ $('Qdrant variables + embedding + KNN neigbours').first().json.qdrantCloudURL }}"}, {"id": "f5612f78-e7d8-4124-9c3a-27bd5870c9bf", "name": "collectionName", "type": "string", "value": "={{ $('Qdrant variables + embedding + KNN neigbours').first().json.collectionName }}"}]}}, "typeVersion": 3.4}, {"id": "8edbff53-cba6-4491-9d5e-bac7ad6db418", "name": "Propagate loop variables", "type": "n8n-nodes-base.set", "position": [640, -240], "parameters": {"options": {}, "assignments": {"assignments": [{"id": "880838bf-2be2-4f5f-9417-974b3cbee163", "name": "=limitKNN", "type": "number", "value": "={{ $json.result.points.length}}"}, {"id": "5fff2bea-f644-4fd9-ad04-afbecd19a5bc", "name": "result", "type": "object", "value": "={{ $json.result }}"}]}}, "typeVersion": 3.4}, {"id": "6fad4cc0-f02c-429d-aa4e-0d69ebab9d65", "name": "Image Test URL", "type": "n8n-nodes-base.set", "position": [-320, -240], "parameters": {"options": {}, "assignments": {"assignments": [{"id": "46ceba40-fb25-450c-8550-d43d8b8aa94c", "name": "imageURL", "type": "string", "value": "={{ $json.query.imageURL }}"}]}}, "typeVersion": 3.4}, {"id": "f02e79e2-32c8-4af0-8bf9-281119b23cc0", "name": "Return class", "type": "n8n-nodes-base.set", "position": [1240, 0], "parameters": {"options": {}, "assignments": {"assignments": [{"id": "bd8ca541-8758-4551-b667-1de373231364", "name": "class", "type": "string", "value": "={{ $json.result[0][0] }}"}]}}, "typeVersion": 3.4}, {"id": "83ca90fb-d5d5-45f4-8957-4363a4baf8ed", "name": "Check tie", "type": "n8n-nodes-base.if", "position": [1040, -240], "parameters": {"options": {}, "conditions": {"options": {"version": 2, "leftValue": "", "caseSensitive": true, "typeValidation": "strict"}, "combinator": "and", "conditions": [{"id": "980663f6-9d7d-4e88-87b9-02030882472c", "operator": {"type": "number", "operation": "gt"}, "leftValue": "={{ $json.result.length }}", "rightValue": 1}, {"id": "9f46fdeb-0f89-4010-99af-624c1c429d6a", "operator": {"type": "number", "operation": "equals"}, "leftValue": "={{ $json.result[0][1] }}", "rightValue": "={{ $json.result[1][1] }}"}, {"id": "c59bc4fe-6821-4639-8595-fdaf4194c1e1", "operator": {"type": "number", "operation": "lte"}, "leftValue": "={{ $('Propagate loop variables').item.json.limitKNN }}", "rightValue": 100}]}}, "typeVersion": 2.2}, {"id": "847ced21-4cfd-45d8-98fa-b578adc054d6", "name": "Qdrant variables + embedding + KNN neigbours", "type": "n8n-nodes-base.set", "position": [120, -240], "parameters": {"options": {}, "assignments": {"assignments": [{"id": "de66070d-5e74-414e-8af7-d094cbc26f62", "name": "ImageEmbedding", "type": "array", "value": "={{ $json.data[0].embedding }}"}, {"id": "58b7384d-fd0c-44aa-9f8e-0306a99be431", "name": "qdrantCloudURL", "type": "string", "value": "=https://152bc6e2-832a-415c-a1aa-fb529f8baf8d.eu-central-1-0.aws.cloud.qdrant.io"}, {"id": "e34c4d88-b102-43cc-a09e-e0553f2da23a", "name": "collectionName", "type": "string", "value": "=land-use"}, {"id": "db37e18d-340b-4624-84f6-df993af866d6", "name": "limitKNN", "type": "number", "value": "=10"}]}}, "typeVersion": 3.4}, {"id": "d1bc4edc-37d2-43ac-8d8b-560453e68d1f", "name": "Sticky Note", "type": "n8n-nodes-base.stickyNote", "position": [-940, -120], "parameters": {"color": 6, "width": 320, "height": 540, "content": "Here we're classifying existing types of satellite imagery of land types:\n- 'agricultural',\n- 'airplane',\n- 'baseballdiamond',\n- 'beach',\n- 'buildings',\n- 'chaparral',\n- 'denseresidential',\n- 'forest',\n- 'freeway',\n- 'golfcourse',\n- 'harbor',\n- 'intersection',\n- 'mediumresidential',\n- 'mobilehomepark',\n- 'overpass',\n- 'parkinglot',\n- 'river',\n- 'runway',\n- 'sparseresidential',\n- 'storagetanks',\n- 'tenniscourt'\n"}, "typeVersion": 1}, {"id": "13560a31-3c72-43b8-9635-3f9ca11f23c9", "name": "Sticky Note1", "type": "n8n-nodes-base.stickyNote", "position": [-520, -460], "parameters": {"color": 6, "content": "I tested this KNN classifier on a whole `test` set of a dataset (it's not a part of the collection, only `validation` + `train` parts). Accuracy of classification on `test` is **93.24%**, no fine-tuning, no metric learning."}, "typeVersion": 1}, {"id": "8c9dcbcb-a1ad-430f-b7dd-e19b5645b0f6", "name": "Execute Workflow Trigger", "type": "n8n-nodes-base.executeWorkflowTrigger", "position": [-520, -240], "parameters": {}, "typeVersion": 1}, {"id": "b36fb270-2101-45e9-bb5c-06c4e07b769c", "name": "Sticky Note2", "type": "n8n-nodes-base.stickyNote", "position": [-1080, -520], "parameters": {"width": 460, "height": 380, "content": "## KNN classification workflow-tool\n### This n8n template takes an image URL (as anomaly detection tool does), and as output, it returns a class of the object on the image (out of land types list)\n\n* An image URL is received via the Execute Workflow Trigger, which is then sent to the Voyage.ai Multimodal Embeddings API to fetch its embedding.\n* The image's embedding vector is then used to query Qdrant, returning a set of X similar images with pre-labeled classes.\n* Majority voting is done for classes of neighbouring images.\n* A loop is used to resolve scenarios where there is a tie in Majority Voting (for example, we have 5 \"forest\" and 5 \"beach\"), and we increase the number of neighbours to retrieve.\n* When the loop finally resolves, the identified class is returned to the calling workflow."}, "typeVersion": 1}, {"id": "51ece7fc-fd85-4d20-ae26-4df2d3893251", "name": "Sticky Note3", "type": "n8n-nodes-base.stickyNote", "position": [120, -40], "parameters": {"height": 200, "content": "Variables define another Qdrant's collection with landscapes (uploaded similarly as the crops collection, don't forget to switch it with your data) + amount of neighbours **limitKNN** in the database we'll use for an input image classification."}, "typeVersion": 1}, {"id": "7aad5904-eb0b-4389-9d47-cc91780737ba", "name": "Sticky Note4", "type": "n8n-nodes-base.stickyNote", "position": [-180, -60], "parameters": {"height": 80, "content": "Similarly to anomaly detection tool, we're embedding input image with the Voyage model"}, "typeVersion": 1}, {"id": "d3702707-ee4a-481f-82ca-d9386f5b7c8a", "name": "Sticky Note5", "type": "n8n-nodes-base.stickyNote", "position": [440, -500], "parameters": {"width": 740, "height": 200, "content": "## Tie loop\nHere we're [querying](https://api.qdrant.tech/api-reference/search/query-points) Qdrant, getting **limitKNN** nearest neighbours to our image <*Query Qdrant node*>, parsing their classes from payloads (images were pre-labeled & uploaded with their labels to Qdrant) & calculating the most frequent class name <*Majority Vote node*>. If there is a tie <*check tie node*> in 2 most common classes, for example, we have 5 \"forest\" and 5 \"harbor\", we repeat the procedure with the number of neighbours increased by 5 <*propagate loop variables node* and *increase limitKNN node*>.\nIf there is no tie, or we have already checked 100 neighbours, we exit the loop <*check tie node*> and return the class-answer."}, "typeVersion": 1}, {"id": "d26911bb-0442-4adc-8511-7cec2d232393", "name": "Sticky Note6", "type": "n8n-nodes-base.stickyNote", "position": [1240, 160], "parameters": {"height": 80, "content": "Here, we extract the name of the input image class decided by the Majority Vote\n"}, "typeVersion": 1}, {"id": "84ffc859-1d5c-4063-9051-3587f30a0017", "name": "Sticky Note10", "type": "n8n-nodes-base.stickyNote", "position": [-520, 80], "parameters": {"color": 4, "width": 540, "height": 260, "content": "### KNN (k nearest neighbours) classification\n1. The first pipeline is uploading (lands) dataset to Qdrant's collection.\n2. **This is the KNN classifier tool, which takes any image as input and classifies it based on queries to the Qdrant (lands) collection.**\n\n### To recreate it\nYou'll have to upload [lands](https://www.kaggle.com/datasets/apollo2506/landuse-scene-classification) dataset from Kaggle to your own Google Storage bucket, and re-create APIs/connections to [Qdrant Cloud](https://qdrant.tech/documentation/quickstart-cloud/) (you can use **Free Tier** cluster), Voyage AI API & Google Cloud Storage\n\n**In general, pipelines are adaptable to any dataset of images**\n"}, "typeVersion": 1}], "active": false, "pinData": {"Execute Workflow Trigger": [{"json": {"query": {"imageURL": "https://storage.googleapis.com/n8n-qdrant-demo/land-use/images_train_test_val/test/buildings/buildings_000323.png"}}}]}, "settings": {"executionOrder": "v1"}, "versionId": "c8cfe732-fd78-4985-9540-ed8cb2de7ef3", "connections": {"Check tie": {"main": [[{"node": "Increase limitKNN", "type": "main", "index": 0}], [{"node": "Return class", "type": "main", "index": 0}]]}, "Embed image": {"main": [[{"node": "Qdrant variables + embedding + KNN neigbours", "type": "main", "index": 0}]]}, "Query Qdrant": {"main": [[{"node": "Propagate loop variables", "type": "main", "index": 0}]]}, "Majority Vote": {"main": [[{"node": "Check tie", "type": "main", "index": 0}]]}, "Image Test URL": {"main": [[{"node": "Embed image", "type": "main", "index": 0}]]}, "Increase limitKNN": {"main": [[{"node": "Query Qdrant", "type": "main", "index": 0}]]}, "Execute Workflow Trigger": {"main": [[{"node": "Image Test URL", "type": "main", "index": 0}]]}, "Propagate loop variables": {"main": [[{"node": "Majority Vote", "type": "main", "index": 0}]]}, "Qdrant variables + embedding + KNN neigbours": {"main": [[{"node": "Query Qdrant", "type": "main", "index": 0}]]}}}