{
  "id": 130500,
  "title": "Solution",
  "content": "<h1>Hardly train timm with wide_resnet50_2 (highly interrupted due to runtime error and overfit also)</h1>\n\n<h1>So trained resnet18, resnet152, resnet50 and found resnet50 suitable most.</h1>\n\n<h1>Adjust num_workers(i.e. no. of cpu's) so to get good results</h1>\n\n<h1>Tried values(num_workers=18,12,6,3)--and found 6 is giving better accuracy also l overfitting the model</h1>\n\n<h1>Increasing image size will increases features , so i train on size=512 (beyond this giving me runtime error)</h1>\n\n<h1>Increase batch size for better accuracy</h1>\n\n<h1>Used transforms like p_affine,p_warp(but not shows such greater impact)</h1>\n\n<h1>Increasing epochs generally lead to overfitting</h1>",
  "comments": [],
  "comment_count": 0,
  "messages": [
    {
      "id": 746034,
      "postDate": "2020-02-14T14:17:22.483Z",
      "content": "<h1>Hardly train timm with wide_resnet50_2 (highly interrupted due to runtime error and overfit also)</h1>\n\n<h1>So trained resnet18, resnet152, resnet50 and found resnet50 suitable most.</h1>\n\n<h1>Adjust num_workers(i.e. no. of cpu's) so to get good results</h1>\n\n<h1>Tried values(num_workers=18,12,6,3)--and found 6 is giving better accuracy also l overfitting the model</h1>\n\n<h1>Increasing image size will increases features , so i train on size=512 (beyond this giving me runtime error)</h1>\n\n<h1>Increase batch size for better accuracy</h1>\n\n<h1>Used transforms like p_affine,p_warp(but not shows such greater impact)</h1>\n\n<h1>Increasing epochs generally lead to overfitting</h1>",
      "rawMarkdown": "# Hardly train timm with wide_resnet50_2 (highly interrupted due to runtime error and overfit also)\n# So trained resnet18, resnet152, resnet50 and found resnet50 suitable most.\n# Adjust num_workers(i.e. no. of cpu's) so to get good results\n# Tried values(num_workers=18,12,6,3)--and found 6 is giving better accuracy also l overfitting the model\n# Increasing image size will increases features , so i train on size=512 (beyond this giving me runtime error)\n# Increase batch size for better accuracy\n# Used transforms like p_affine,p_warp(but not shows such greater impact)\n# Increasing epochs generally lead to overfitting "
    }
  ],
  "complete": true,
  "source": "live",
  "fetched_at": "2026-10-20T05:03:30Z",
  "competition_id": 18191,
  "competition_slug": "cl-cv2020",
  "team_id": 4288974,
  "solution_url": "/competitions/cl-cv2020/writeups/ai-fanatics-solution"
}