Here are link on two files (train_index.7z and test_index.7z). Both files are in JSON format. Files contain TOP50 words described each picture. Words were extracted with MXNet on pretrained model Inception-V3 Network. Words oredered by it's importance for given picture.
\n\nFiles have the following format:
\n\n{"267142.jpg": {"17": "630", "10": "953", "25": "585", "46": "234", "36": "704", "29": "673", "49": "369", "41": "50", "32": "373", "26": "873", "13": "830", "22": "742", "28": "672", "1": "806", "5": "900", "6": "730", "19": "694", "39": "891", "2": "738", "11": "740", "3": "754", "35": "613", "24": "737", "43": "402", "0": "734", "48": "617", "44": "829", "16": "688", "8": "974", "15": "812", "31": "618", "34": "655", "45": "318", "9": "676", "23": "362", "47": "455", "30": "375", "12": "331", "40": "224", "27": "999", "38": "736", "7": "793", "4": "372", "18": "329", "14": "864", "42": "197", "37": "386", "20": "948", "33": "766", "21": "616"}, "307998.jpg": {"17": "808", "7": "806", "10": "892", "25": "188", "46": "591", "36": "914", "29": "878", "49": "637", "41": "644", "32": "864", "26": "229", "13": "318", "22": "736", "11": "859", "1": "844", "5": "793", "28": "397", "3": "823", "6": "754", "19": "768", "39": "581", "2": "734", "27": "230", "35": "377", "12": "829", "0": "974", "47": "472", "48": "501", "44": "461", "16": "947", "8": "771", "15": "824", "31": "283", "34": "362", "45": "386", "9": "822", "23": "525", "33": "284", "30": "873", "43": "129", "40": "738", "38": "813", "24": "694", "4": "953", "18": "218", "14": "838", "42": "577", "37": "213", "20": "980", "21": "73"}, "1.jpg" ... }
\n\nHere:
\n\n267142.jpg - file name of foto
\n\n"17": "630" - "630" is WORD_ID, "17" is place in ordered list of words describing this picture. Index "0" contains the most important word describing picture, "49" contains the less important word.
\n\nMap of WORD_ID to English words can be found in file synset.txt.
", "messages": [ { "id": "112133", "postDate": "03/18/2016 09:42:33", "content": "Here are link on two files (train_index.7z and test_index.7z). Both files are in JSON format. Files contain TOP50 words described each picture. Words were extracted with MXNet on pretrained model Inception-V3 Network. Words oredered by it's importance for given picture.
\n\nFiles have the following format:
\n\n{"267142.jpg": {"17": "630", "10": "953", "25": "585", "46": "234", "36": "704", "29": "673", "49": "369", "41": "50", "32": "373", "26": "873", "13": "830", "22": "742", "28": "672", "1": "806", "5": "900", "6": "730", "19": "694", "39": "891", "2": "738", "11": "740", "3": "754", "35": "613", "24": "737", "43": "402", "0": "734", "48": "617", "44": "829", "16": "688", "8": "974", "15": "812", "31": "618", "34": "655", "45": "318", "9": "676", "23": "362", "47": "455", "30": "375", "12": "331", "40": "224", "27": "999", "38": "736", "7": "793", "4": "372", "18": "329", "14": "864", "42": "197", "37": "386", "20": "948", "33": "766", "21": "616"}, "307998.jpg": {"17": "808", "7": "806", "10": "892", "25": "188", "46": "591", "36": "914", "29": "878", "49": "637", "41": "644", "32": "864", "26": "229", "13": "318", "22": "736", "11": "859", "1": "844", "5": "793", "28": "397", "3": "823", "6": "754", "19": "768", "39": "581", "2": "734", "27": "230", "35": "377", "12": "829", "0": "974", "47": "472", "48": "501", "44": "461", "16": "947", "8": "771", "15": "824", "31": "283", "34": "362", "45": "386", "9": "822", "23": "525", "33": "284", "30": "873", "43": "129", "40": "738", "38": "813", "24": "694", "4": "953", "18": "218", "14": "838", "42": "577", "37": "213", "20": "980", "21": "73"}, "1.jpg" ... }
\n\nHere:
\n\n267142.jpg - file name of foto
\n\n"17": "630" - "630" is WORD_ID, "17" is place in ordered list of words describing this picture. Index "0" contains the most important word describing picture, "49" contains the less important word.
\n\nMap of WORD_ID to English words can be found in file synset.txt.
", "rawMarkdown": "Here are link on two files (train_index.7z and test_index.7z). Both files are in JSON format. Files contain TOP50 words described each picture. Words were extracted with MXNet on pretrained model [Inception-V3 Network][1]. Words oredered by it's importance for given picture.\r\n\r\nFiles have the following format:\r\n\r\n {\"267142.jpg\": {\"17\": \"630\", \"10\": \"953\", \"25\": \"585\", \"46\": \"234\", \"36\": \"704\", \"29\": \"673\", \"49\": \"369\", \"41\": \"50\", \"32\": \"373\", \"26\": \"873\", \"13\": \"830\", \"22\": \"742\", \"28\": \"672\", \"1\": \"806\", \"5\": \"900\", \"6\": \"730\", \"19\": \"694\", \"39\": \"891\", \"2\": \"738\", \"11\": \"740\", \"3\": \"754\", \"35\": \"613\", \"24\": \"737\", \"43\": \"402\", \"0\": \"734\", \"48\": \"617\", \"44\": \"829\", \"16\": \"688\", \"8\": \"974\", \"15\": \"812\", \"31\": \"618\", \"34\": \"655\", \"45\": \"318\", \"9\": \"676\", \"23\": \"362\", \"47\": \"455\", \"30\": \"375\", \"12\": \"331\", \"40\": \"224\", \"27\": \"999\", \"38\": \"736\", \"7\": \"793\", \"4\": \"372\", \"18\": \"329\", \"14\": \"864\", \"42\": \"197\", \"37\": \"386\", \"20\": \"948\", \"33\": \"766\", \"21\": \"616\"}, \"307998.jpg\": {\"17\": \"808\", \"7\": \"806\", \"10\": \"892\", \"25\": \"188\", \"46\": \"591\", \"36\": \"914\", \"29\": \"878\", \"49\": \"637\", \"41\": \"644\", \"32\": \"864\", \"26\": \"229\", \"13\": \"318\", \"22\": \"736\", \"11\": \"859\", \"1\": \"844\", \"5\": \"793\", \"28\": \"397\", \"3\": \"823\", \"6\": \"754\", \"19\": \"768\", \"39\": \"581\", \"2\": \"734\", \"27\": \"230\", \"35\": \"377\", \"12\": \"829\", \"0\": \"974\", \"47\": \"472\", \"48\": \"501\", \"44\": \"461\", \"16\": \"947\", \"8\": \"771\", \"15\": \"824\", \"31\": \"283\", \"34\": \"362\", \"45\": \"386\", \"9\": \"822\", \"23\": \"525\", \"33\": \"284\", \"30\": \"873\", \"43\": \"129\", \"40\": \"738\", \"38\": \"813\", \"24\": \"694\", \"4\": \"953\", \"18\": \"218\", \"14\": \"838\", \"42\": \"577\", \"37\": \"213\", \"20\": \"980\", \"21\": \"73\"}, \"1.jpg\" ... }\r\n\r\nHere: \r\n\r\n**267142.jpg** - file name of foto\r\n\r\n**\"17\": \"630\"** - \"630\" is WORD_ID, \"17\" is place in ordered list of words describing this picture. Index \"0\" contains the most important word describing picture, \"49\" contains the less important word.\r\n\r\nMap of WORD_ID to English words can be found in file **synset.txt**. \r\n\r\n\r\n [1]: https://github.com/dmlc/mxnet-model-gallery/blob/master/imagenet-1k-inception-v3.md", "votes": null }, { "id": "112255", "postDate": "03/19/2016 06:40:23", "content": "I decided to post the code to generate these files: code on GITHub.
\n\nTo run it you need:
\n\n1) Copy all files in model directory from Inception repository.
\n\n2) You need to run this code twice for ../input/train_photos folder and for ../input/test_photos folder. Just change run_stage = 1 variable at line 39 for 2nd stage.
\n\n3) In case you have MXNet compiled with GPU, change ctx=mx.cpu() to ctx=mx.gpu() on line 12.
\n\nIn CPU mode it requires around a week to complete.
", "rawMarkdown": "I decided to post the code to generate these files: [code on GITHub][1].\r\n\r\nTo run it you need:\r\n\r\n1) Copy all files in **model** directory from [Inception repository][2].\r\n\r\n2) You need to run this code twice for ../input/train_photos folder and for ../input/test_photos folder. Just change **run_stage = 1** variable at line 39 for 2nd stage.\r\n\r\n3) In case you have MXNet compiled with GPU, change **ctx=mx.cpu()** to **ctx=mx.gpu()** on line 12.\r\n\r\nIn CPU mode it requires around a week to complete.\r\n\r\n [1]: https://github.com/ZFTurbo/KAGGLE_YELP/blob/master/run-inception-code.py\r\n [2]: https://github.com/dmlc/mxnet-model-gallery/blob/master/imagenet-1k-inception-v3.md", "votes": null }, { "id": "112506", "postDate": "03/21/2016 16:29:17", "content": "Thanks for sharing that! Question: How can I retrieve the actual word from the synset? The words within the JSON are numbered from 0 to 999, but the synset words have identifiers such as "n02119789".
", "rawMarkdown": "Thanks for sharing that! Question: How can I retrieve the actual word from the synset? The words within the JSON are numbered from 0 to 999, but the synset words have identifiers such as \"n02119789\".", "votes": null }, { "id": "112523", "postDate": "03/21/2016 18:00:57", "content": "WORD_ID is the line number in synset.txt file.
", "rawMarkdown": "WORD_ID is the line number in **synset.txt** file.", "votes": null }, { "id": "112528", "postDate": "03/21/2016 18:20:01", "content": "Thx for clarifying!
", "rawMarkdown": "Thx for clarifying!", "votes": null } ], "comments": [ { "id": 112255, "author_name": "zfturbo", "author_url": "", "post_date": "03/19/2016 06:40:23", "content": "I decided to post the code to generate these files: code on GITHub.
\n\nTo run it you need:
\n\n1) Copy all files in model directory from Inception repository.
\n\n2) You need to run this code twice for ../input/train_photos folder and for ../input/test_photos folder. Just change run_stage = 1 variable at line 39 for 2nd stage.
\n\n3) In case you have MXNet compiled with GPU, change ctx=mx.cpu() to ctx=mx.gpu() on line 12.
\n\nIn CPU mode it requires around a week to complete.
", "votes": null, "replies": [] }, { "id": 112506, "author_name": "qqilihq", "author_url": "", "post_date": "03/21/2016 16:29:17", "content": "Thanks for sharing that! Question: How can I retrieve the actual word from the synset? The words within the JSON are numbered from 0 to 999, but the synset words have identifiers such as "n02119789".
", "votes": null, "replies": [] }, { "id": 112523, "author_name": "zfturbo", "author_url": "", "post_date": "03/21/2016 18:00:57", "content": "WORD_ID is the line number in synset.txt file.
", "votes": null, "replies": [] }, { "id": 112528, "author_name": "qqilihq", "author_url": "", "post_date": "03/21/2016 18:20:01", "content": "Thx for clarifying!
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "112133": "Here are link on two files (train_index.7z and test_index.7z). Both files are in JSON format. Files contain TOP50 words described each picture. Words were extracted with MXNet on pretrained model [Inception-V3 Network][1]. Words oredered by it's importance for given picture.\r\n\r\nFiles have the following format:\r\n\r\n {\"267142.jpg\": {\"17\": \"630\", \"10\": \"953\", \"25\": \"585\", \"46\": \"234\", \"36\": \"704\", \"29\": \"673\", \"49\": \"369\", \"41\": \"50\", \"32\": \"373\", \"26\": \"873\", \"13\": \"830\", \"22\": \"742\", \"28\": \"672\", \"1\": \"806\", \"5\": \"900\", \"6\": \"730\", \"19\": \"694\", \"39\": \"891\", \"2\": \"738\", \"11\": \"740\", \"3\": \"754\", \"35\": \"613\", \"24\": \"737\", \"43\": \"402\", \"0\": \"734\", \"48\": \"617\", \"44\": \"829\", \"16\": \"688\", \"8\": \"974\", \"15\": \"812\", \"31\": \"618\", \"34\": \"655\", \"45\": \"318\", \"9\": \"676\", \"23\": \"362\", \"47\": \"455\", \"30\": \"375\", \"12\": \"331\", \"40\": \"224\", \"27\": \"999\", \"38\": \"736\", \"7\": \"793\", \"4\": \"372\", \"18\": \"329\", \"14\": \"864\", \"42\": \"197\", \"37\": \"386\", \"20\": \"948\", \"33\": \"766\", \"21\": \"616\"}, \"307998.jpg\": {\"17\": \"808\", \"7\": \"806\", \"10\": \"892\", \"25\": \"188\", \"46\": \"591\", \"36\": \"914\", \"29\": \"878\", \"49\": \"637\", \"41\": \"644\", \"32\": \"864\", \"26\": \"229\", \"13\": \"318\", \"22\": \"736\", \"11\": \"859\", \"1\": \"844\", \"5\": \"793\", \"28\": \"397\", \"3\": \"823\", \"6\": \"754\", \"19\": \"768\", \"39\": \"581\", \"2\": \"734\", \"27\": \"230\", \"35\": \"377\", \"12\": \"829\", \"0\": \"974\", \"47\": \"472\", \"48\": \"501\", \"44\": \"461\", \"16\": \"947\", \"8\": \"771\", \"15\": \"824\", \"31\": \"283\", \"34\": \"362\", \"45\": \"386\", \"9\": \"822\", \"23\": \"525\", \"33\": \"284\", \"30\": \"873\", \"43\": \"129\", \"40\": \"738\", \"38\": \"813\", \"24\": \"694\", \"4\": \"953\", \"18\": \"218\", \"14\": \"838\", \"42\": \"577\", \"37\": \"213\", \"20\": \"980\", \"21\": \"73\"}, \"1.jpg\" ... }\r\n\r\nHere: \r\n\r\n**267142.jpg** - file name of foto\r\n\r\n**\"17\": \"630\"** - \"630\" is WORD_ID, \"17\" is place in ordered list of words describing this picture. Index \"0\" contains the most important word describing picture, \"49\" contains the less important word.\r\n\r\nMap of WORD_ID to English words can be found in file **synset.txt**. \r\n\r\n\r\n [1]: https://github.com/dmlc/mxnet-model-gallery/blob/master/imagenet-1k-inception-v3.md", "112255": "I decided to post the code to generate these files: [code on GITHub][1].\r\n\r\nTo run it you need:\r\n\r\n1) Copy all files in **model** directory from [Inception repository][2].\r\n\r\n2) You need to run this code twice for ../input/train_photos folder and for ../input/test_photos folder. Just change **run_stage = 1** variable at line 39 for 2nd stage.\r\n\r\n3) In case you have MXNet compiled with GPU, change **ctx=mx.cpu()** to **ctx=mx.gpu()** on line 12.\r\n\r\nIn CPU mode it requires around a week to complete.\r\n\r\n [1]: https://github.com/ZFTurbo/KAGGLE_YELP/blob/master/run-inception-code.py\r\n [2]: https://github.com/dmlc/mxnet-model-gallery/blob/master/imagenet-1k-inception-v3.md", "112506": "Thanks for sharing that! Question: How can I retrieve the actual word from the synset? The words within the JSON are numbered from 0 to 999, but the synset words have identifiers such as \"n02119789\".", "112523": "WORD_ID is the line number in **synset.txt** file.", "112528": "Thx for clarifying!" }, "source": "meta" }