{ "id": 149641, "title": "Where is metric code reference material", "url": "/competitions/imaterialist-fashion-2020-fgvc7/discussion/149641", "author_name": "fan", "post_date": "2020-05-09T09:41:34.364000", "votes": 1, "comment_count": 8, "views": 0, "content": "

hey @makeitworkjml, check this out

\n\n

My score is always zero, I need a metric code to evaluate my model, May I have some suggestions~

", "messages": [ { "id": 845670, "postDate": "2020-05-13T11:07:56.753Z", "content": "

I use pycocotools , segmentation is too bad .

\n\n

loading annotations into memory...\nDone (t=0.00s)\ncreating index...\nindex created!\nLoading and preparing results...\nDONE (t=0.00s)\ncreating index...\nindex created!\nRunning per image evaluation...\nEvaluate annotation type segm\nDONE (t=0.03s).\nAccumulating evaluation results...\nDONE (t=0.07s).\n Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.010\n Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.026\n Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.007\n Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000\n Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.024\n Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.011\n Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.017\n Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.017\n Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.017\n Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000\n Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.047\n Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.015

", "rawMarkdown": "I use pycocotools , segmentation is too bad . \n\n\nloading annotations into memory...\nDone (t=0.00s)\ncreating index...\nindex created!\nLoading and preparing results...\nDONE (t=0.00s)\ncreating index...\nindex created!\nRunning per image evaluation...\nEvaluate annotation type *segm*\nDONE (t=0.03s).\nAccumulating evaluation results...\nDONE (t=0.07s).\n Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.010\n Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.026\n Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.007\n Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000\n Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.024\n Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.011\n Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.017\n Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.017\n Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.017\n Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000\n Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.047\n Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.015", "votes": 1 }, { "id": 839318, "postDate": "2020-05-09T09:41:34.363Z", "content": "

hey @makeitworkjml, check this out

\n\n

My score is always zero, I need a metric code to evaluate my model, May I have some suggestions~

", "rawMarkdown": "

hey @makeitworkjml, check this out

\n\nMy score is always zero, I need a metric code to evaluate my model, May I have some suggestions~", "votes": 1 }, { "id": 851993, "postDate": "2020-05-18T05:13:56.017Z", "content": "

clothing items has more complex contours and silhouette, so to improve segmentation is definitely one direction!

", "rawMarkdown": "clothing items has more complex contours and silhouette, so to improve segmentation is definitely one direction!", "replies": [ { "id": 852262, "postDate": "2020-05-18T09:37:42.577Z", "content": "

Thanks~

", "rawMarkdown": "Thanks~" }, { "id": 853404, "postDate": "2020-05-19T06:45:40.223Z", "content": "

I'm curious to see how to improve the segmentation masks! Please feel free to share your work after the competition. :)

", "rawMarkdown": "I'm curious to see how to improve the segmentation masks! Please feel free to share your work after the competition. :) " }, { "id": 862972, "postDate": "2020-05-27T01:54:23.733Z", "content": "

OK，I will try it , :)

", "rawMarkdown": "OK，I will try it , :) " }, { "id": 864863, "postDate": "2020-05-28T08:30:05.407Z", "content": "

I use the MaskRCNN from : https://github.com/matterport/Mask_RCNN\nAnd set: \n\n IMAGES_PER_GPU = 1\n

\n\n

Enlarge image size:
\n\n # Use small images for faster training. Set the limits of the small side\n # the large side, and that determines the image shape.\n IMAGE_MIN_DIM = 1536\n IMAGE_MAX_DIM = 1536\n

\n\n

Anchor K-Means:
\n\n # Length of square anchor side in pixels\n RPN_ANCHOR_SCALES = (75, 187, 308, 554, 1032)\n # Ratios of anchors at each cell (width/height)\n # A value of 1 represents a square anchor, and 0.5 is a wide anchor\n RPN_ANCHOR_RATIOS = [0.7, 0.9, 1.2] \n

\n\n

Enlarge mask resolution:\n\n MASK_POOL_SIZE = 28 \n MASK_SHAPE = [56, 56]\n

", "rawMarkdown": "I use the MaskRCNN from : https://github.com/matterport/Mask_RCNN\nAnd set: \n```\n IMAGES_PER_GPU = 1\n```\n\nEnlarge image size: \n``` \n # Use small images for faster training. Set the limits of the small side\n # the large side, and that determines the image shape.\n IMAGE_MIN_DIM = 1536\n IMAGE_MAX_DIM = 1536\n```\n\nAnchor K-Means: \n```\n # Length of square anchor side in pixels\n RPN_ANCHOR_SCALES = (75, 187, 308, 554, 1032)\n # Ratios of anchors at each cell (width/height)\n # A value of 1 represents a square anchor, and 0.5 is a wide anchor\n RPN_ANCHOR_RATIOS = [0.7, 0.9, 1.2] \n```\n\nEnlarge mask resolution:\n```\n MASK_POOL_SIZE = 28 \n MASK_SHAPE = [56, 56]\n```" }, { "id": 927312, "postDate": "2020-07-13T10:20:59.787Z", "rawMarkdown": "", "isDeleted": true }, { "id": 932416, "postDate": "2020-07-17T03:48:31.520Z", "content": "

@deeeeeeeep We are preparing to release the training code this week. Also, you can checkout the fashionpedia-api for more comprehensive evaluation metrics than what we have here.

", "rawMarkdown": "@deeeeeeeep We are preparing to release the training code this week. Also, you can checkout the [fashionpedia-api ](https://github.com/KMnP/fashionpedia-api)for more comprehensive evaluation metrics than what we have here." } ] } ], "comments": [ { "id": 845670, "author_name": "fan", "author_url": "", "post_date": "2020-05-13T11:07:56.753000", "content": "

I use pycocotools , segmentation is too bad .

\n\n

loading annotations into memory...\nDone (t=0.00s)\ncreating index...\nindex created!\nLoading and preparing results...\nDONE (t=0.00s)\ncreating index...\nindex created!\nRunning per image evaluation...\nEvaluate annotation type segm\nDONE (t=0.03s).\nAccumulating evaluation results...\nDONE (t=0.07s).\n Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.010\n Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.026\n Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.007\n Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000\n Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.024\n Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.011\n Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.017\n Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.017\n Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.017\n Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000\n Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.047\n Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.015

", "votes": 1, "replies": [] }, { "id": 851993, "author_name": "Menglin Jia", "author_url": "", "post_date": "2020-05-18T05:13:56.017000", "content": "

clothing items has more complex contours and silhouette, so to improve segmentation is definitely one direction!

", "votes": 0, "replies": [ { "id": 852262, "author_name": "fan", "author_url": "", "post_date": "2020-05-18T09:37:42.577000", "content": "

Thanks~

", "votes": 0, "replies": [] }, { "id": 853404, "author_name": "Menglin Jia", "author_url": "", "post_date": "2020-05-19T06:45:40.223000", "content": "

I'm curious to see how to improve the segmentation masks! Please feel free to share your work after the competition. :)

", "votes": 0, "replies": [] }, { "id": 862972, "author_name": "fan", "author_url": "", "post_date": "2020-05-27T01:54:23.733000", "content": "

OK，I will try it , :)

", "votes": 0, "replies": [] }, { "id": 864863, "author_name": "fan", "author_url": "", "post_date": "2020-05-28T08:30:05.407000", "content": "

I use the MaskRCNN from : https://github.com/matterport/Mask_RCNN\nAnd set: \n\n IMAGES_PER_GPU = 1\n

\n\n

Enlarge image size:
\n\n # Use small images for faster training. Set the limits of the small side\n # the large side, and that determines the image shape.\n IMAGE_MIN_DIM = 1536\n IMAGE_MAX_DIM = 1536\n

\n\n

Anchor K-Means:
\n\n # Length of square anchor side in pixels\n RPN_ANCHOR_SCALES = (75, 187, 308, 554, 1032)\n # Ratios of anchors at each cell (width/height)\n # A value of 1 represents a square anchor, and 0.5 is a wide anchor\n RPN_ANCHOR_RATIOS = [0.7, 0.9, 1.2] \n

\n\n

Enlarge mask resolution:\n\n MASK_POOL_SIZE = 28 \n MASK_SHAPE = [56, 56]\n

", "votes": 0, "replies": [] }, { "id": 927312, "author_name": "", "author_url": "", "post_date": "2020-07-13T10:20:59.787000", "content": "", "votes": 0, "replies": [] }, { "id": 932416, "author_name": "Menglin Jia", "author_url": "", "post_date": "2020-07-17T03:48:31.520000", "content": "

@deeeeeeeep We are preparing to release the training code this week. Also, you can checkout the fashionpedia-api for more comprehensive evaluation metrics than what we have here.

", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "845670": "I use pycocotools , segmentation is too bad . \n\n\nloading annotations into memory...\nDone (t=0.00s)\ncreating index...\nindex created!\nLoading and preparing results...\nDONE (t=0.00s)\ncreating index...\nindex created!\nRunning per image evaluation...\nEvaluate annotation type *segm*\nDONE (t=0.03s).\nAccumulating evaluation results...\nDONE (t=0.07s).\n Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.010\n Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.026\n Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.007\n Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000\n Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.024\n Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.011\n Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.017\n Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.017\n Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.017\n Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000\n Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.047\n Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.015", "839318": "

hey @makeitworkjml, check this out

\n\nMy score is always zero, I need a metric code to evaluate my model, May I have some suggestions~", "851993": "clothing items has more complex contours and silhouette, so to improve segmentation is definitely one direction!" } }