I encounter submission error with this notebook.\nI am very confused because I have the right predicted bounding boxes there.\nBelow is my notebook:\nhttps://www.kaggle.com/chrisstan/yolov4-inference?scriptVersionId=39017659
", "messages": [ { "id": "934506", "postDate": "07/18/2020 13:54:44", "content": "I encounter submission error with this notebook.\nI am very confused because I have the right predicted bounding boxes there.\nBelow is my notebook:\nhttps://www.kaggle.com/chrisstan/yolov4-inference?scriptVersionId=39017659
", "rawMarkdown": "I encounter submission error with this notebook.\nI am very confused because I have the right predicted bounding boxes there.\nBelow is my notebook:\nhttps://www.kaggle.com/chrisstan/yolov4-inference?scriptVersionId=39017659", "votes": null }, { "id": "935492", "postDate": "07/19/2020 12:42:42", "content": "Not a solution but I was looking into your notebook and noticed although you have set your YoloV4 weights dataset private you copy those in the output folder resulting in it becoming publicly visible/available.
\n\nIt would be a shame if some free loaders took advantage of this mistake. Please look into this issue and maybe setup a new revised notebook.
", "rawMarkdown": "Not a solution but I was looking into your notebook and noticed although you have set your YoloV4 weights dataset private you copy those in the output folder resulting in it becoming publicly visible/available.\n\nIt would be a shame if some free loaders took advantage of this mistake. Please look into this issue and maybe setup a new revised notebook.", "votes": null }, { "id": "935990", "postDate": "07/19/2020 21:44:05", "content": "Thank for your suggestion.\nI have fixed it.
", "rawMarkdown": "Thank for your suggestion.\nI have fixed it.", "votes": null }, { "id": "936270", "postDate": "07/20/2020 05:43:12", "content": "Your code looks similar to this so follow below solution-
\nhttps://www.kaggle.com/c/global-wheat-detection/discussion/168179
I tried like this and still failed..\n```python\nresults = []
\n\nfor images, image_ids in data_loader:\n predictions = make_tta_predictions(images)\n try:\n for i, image in enumerate(images):
\n\n boxes, scores, labels = run_wbf(predictions, image_index=i)\n boxes = (boxes*(1024 / 704)).round().astype(np.int32).clip(min=0, max=1023)\n image_id = image_ids[i]\n\n boxes[:, 2] = boxes[:, 2] - boxes[:, 0]\n boxes[:, 3] = boxes[:, 3] - boxes[:, 1]\n\n result = {\n 'image_id': image_id,\n 'PredictionString': format_prediction_string(boxes, scores)\n }\n results.append(result)\nexcept:\n result = {\n 'image_id': image_id,\n 'PredictionString': ''\n }\n results.append(result)\n```
", "rawMarkdown": "I tried like this and still failed..\n```python\nresults = []\n\nfor images, image_ids in data_loader:\n predictions = make_tta_predictions(images)\n try:\n for i, image in enumerate(images):\n\n boxes, scores, labels = run_wbf(predictions, image_index=i)\n boxes = (boxes*(1024 / 704)).round().astype(np.int32).clip(min=0, max=1023)\n image_id = image_ids[i]\n\n boxes[:, 2] = boxes[:, 2] - boxes[:, 0]\n boxes[:, 3] = boxes[:, 3] - boxes[:, 1]\n\n result = {\n 'image_id': image_id,\n 'PredictionString': format_prediction_string(boxes, scores)\n }\n results.append(result)\n except:\n result = {\n 'image_id': image_id,\n 'PredictionString': ''\n }\n results.append(result)\n```", "votes": null }, { "id": "936856", "postDate": "07/20/2020 14:59:57", "content": "Check box length also like below-
\nif len(boxes)>0:\n result = {\n 'image_id': images_info['filename'][:-4],\n 'PredictionString': format_prediction_string(boxes, scores)}\n else:\n result = {\n 'image_id': images_info['filename'][:-4],\n 'PredictionString': ''}\n results.append(result)
And also put your predictions = make_tta_predictions(images) inside the try block.
Just a wild guess but some non 704 sized image maybe causing an error in your make_tta_predictions(images)\nWe've too screwed a lot of our submissions over such rogue images which managed to reach the model input through some loophole in our code.
Try copying all the test images (~1200 during submission) from the input folder to the output folder (kaggle/working/) and resize them all to your desired image size (704) right away.\nFeed this new test directory to your test loader function and see if the error still persists.
\n\nOr just put a lazy condition like \nif image.shape[0] != 704 or image.shape[1] != 704: image=cv2.resize(image, (704,704))\npredictions = make_tta_predictions(images)
You can also locally test this theory of mine by taking some ~1000 train images as test images with weird shapes (take a few greater than 1024 some lesser than it and maybe try rectangle images etc.) and run your inference with them.
\n\nSee if this helps.
", "rawMarkdown": "Just a wild guess but some non 704 sized image maybe causing an error in your `make_tta_predictions(images)`\nWe've too screwed a lot of our submissions over such rogue images which managed to reach the model input through some loophole in our code.\n\nTry copying all the test images (~1200 during submission) from the input folder to the output folder (kaggle/working/) and resize them all to your desired image size (704) right away.\nFeed this new test directory to your test loader function and see if the error still persists.\n\nOr just put a lazy condition like \n`if image.shape[0] != 704 or image.shape[1] != 704: image=cv2.resize(image, (704,704))\npredictions = make_tta_predictions(images)`\n\nYou can also locally test this theory of mine by taking some ~1000 train images as test images with weird shapes (take a few greater than 1024 some lesser than it and maybe try rectangle images etc.) and run your inference with them.\n\nSee if this helps.", "votes": null }, { "id": "937200", "postDate": "07/20/2020 19:56:21", "content": "Hi, which submission error are you getting? I can only speak from my experience:\n1. First I got an \"Submission CSV not found\" error. This seems to have been caused by me having an additional folder in my working directory, so now I delete everything except submission.csv and the notebook. It could also have been caused by me copying too much data into the working directory, because I was copying the whole training set. I think that might have been too much (first competition errors).\n2. I then got a \"Submission Scoring Error\". To solve this I first added a cell copying sample_submission.csv as submision.csv into the working directory --> no error, so I knew that the error was not caused by my code and that the system also accepts incomplete submissions without error.\nThen I did the same for a submission made by my code on only the 10 public images --> error. So I knew that the error was caused by the formating of the file containing only the 10 public images. After schecking everything I could think of (double spaces, etc.), I went to https://regex101.com/ copied in the whole submission file and used the regex r\"0.\\d+ \\d+ \\d+ \\d+ \\d+\" to match all of my output. I saw that a few little parts did not match because the submission did not contain all 4 coordinates, only 3 in some rare cases. I then corrected the problem and it works now.\nHope this helps!\nEdit: I saw that you already fixed it. But maybe someone else will find this.
", "rawMarkdown": "Hi, which submission error are you getting? I can only speak from my experience:\n1. First I got an \"Submission CSV not found\" error. This seems to have been caused by me having an additional folder in my working directory, so now I delete everything except submission.csv and the notebook. It could also have been caused by me copying too much data into the working directory, because I was copying the whole training set. I think that might have been too much (first competition errors).\n2. I then got a \"Submission Scoring Error\". To solve this I first added a cell copying sample_submission.csv as submision.csv into the working directory --> no error, so I knew that the error was not caused by my code and that the system also accepts incomplete submissions without error.\nThen I did the same for a submission made by my code on only the 10 public images --> error. So I knew that the error was caused by the formating of the file containing only the 10 public images. After schecking everything I could think of (double spaces, etc.), I went to https://regex101.com/ copied in the whole submission file and used the regex r\"0.\\d+ \\d+ \\d+ \\d+ \\d+\" to match all of my output. I saw that a few little parts did not match because the submission did not contain all 4 coordinates, only 3 in some rare cases. I then corrected the problem and it works now.\nHope this helps!\nEdit: I saw that you already fixed it. But maybe someone else will find this.", "votes": null }, { "id": "937320", "postDate": "07/20/2020 23:39:17", "content": "Now it can submit! Thanks!
", "rawMarkdown": "Now it can submit! Thanks!", "votes": null }, { "id": "937321", "postDate": "07/20/2020 23:39:32", "content": "Now it can work! Thanks!!!
", "rawMarkdown": "Now it can work! Thanks!!!", "votes": null }, { "id": "937330", "postDate": "07/20/2020 23:56:42", "content": "Now it work! Thanks !!!!!
", "rawMarkdown": "Now it work! Thanks !!!!!", "votes": null } ], "comments": [ { "id": 936270, "author_name": "iamprateek", "author_url": "", "post_date": "07/20/2020 05:43:12", "content": "Your code looks similar to this so follow below solution-
\nhttps://www.kaggle.com/c/global-wheat-detection/discussion/168179
I tried like this and still failed..\n```python\nresults = []
\n\nfor images, image_ids in data_loader:\n predictions = make_tta_predictions(images)\n try:\n for i, image in enumerate(images):
\n\n boxes, scores, labels = run_wbf(predictions, image_index=i)\n boxes = (boxes*(1024 / 704)).round().astype(np.int32).clip(min=0, max=1023)\n image_id = image_ids[i]\n\n boxes[:, 2] = boxes[:, 2] - boxes[:, 0]\n boxes[:, 3] = boxes[:, 3] - boxes[:, 1]\n\n result = {\n 'image_id': image_id,\n 'PredictionString': format_prediction_string(boxes, scores)\n }\n results.append(result)\nexcept:\n result = {\n 'image_id': image_id,\n 'PredictionString': ''\n }\n results.append(result)\n```
", "votes": null, "replies": [ { "id": 936856, "author_name": "iamprateek", "author_url": "", "post_date": "07/20/2020 14:59:57", "content": "Check box length also like below-
\nif len(boxes)>0:\n result = {\n 'image_id': images_info['filename'][:-4],\n 'PredictionString': format_prediction_string(boxes, scores)}\n else:\n result = {\n 'image_id': images_info['filename'][:-4],\n 'PredictionString': ''}\n results.append(result)
And also put your predictions = make_tta_predictions(images) inside the try block.
Now it can work! Thanks!!!
", "votes": null, "replies": [] } ] }, { "id": 936864, "author_name": "ii5m0k3ii", "author_url": "", "post_date": "07/20/2020 15:03:23", "content": "Just a wild guess but some non 704 sized image maybe causing an error in your make_tta_predictions(images)\nWe've too screwed a lot of our submissions over such rogue images which managed to reach the model input through some loophole in our code.
Try copying all the test images (~1200 during submission) from the input folder to the output folder (kaggle/working/) and resize them all to your desired image size (704) right away.\nFeed this new test directory to your test loader function and see if the error still persists.
\n\nOr just put a lazy condition like \nif image.shape[0] != 704 or image.shape[1] != 704: image=cv2.resize(image, (704,704))\npredictions = make_tta_predictions(images)
You can also locally test this theory of mine by taking some ~1000 train images as test images with weird shapes (take a few greater than 1024 some lesser than it and maybe try rectangle images etc.) and run your inference with them.
\n\nSee if this helps.
", "votes": null, "replies": [ { "id": 937320, "author_name": "chrisstan", "author_url": "", "post_date": "07/20/2020 23:39:17", "content": "Now it can submit! Thanks!
", "votes": null, "replies": [] } ] }, { "id": 935492, "author_name": "ii5m0k3ii", "author_url": "", "post_date": "07/19/2020 12:42:42", "content": "Not a solution but I was looking into your notebook and noticed although you have set your YoloV4 weights dataset private you copy those in the output folder resulting in it becoming publicly visible/available.
\n\nIt would be a shame if some free loaders took advantage of this mistake. Please look into this issue and maybe setup a new revised notebook.
", "votes": null, "replies": [ { "id": 935990, "author_name": "chrisstan", "author_url": "", "post_date": "07/19/2020 21:44:05", "content": "Thank for your suggestion.\nI have fixed it.
", "votes": null, "replies": [] } ] }, { "id": 937200, "author_name": "johannesillerhaus", "author_url": "", "post_date": "07/20/2020 19:56:21", "content": "Hi, which submission error are you getting? I can only speak from my experience:\n1. First I got an \"Submission CSV not found\" error. This seems to have been caused by me having an additional folder in my working directory, so now I delete everything except submission.csv and the notebook. It could also have been caused by me copying too much data into the working directory, because I was copying the whole training set. I think that might have been too much (first competition errors).\n2. I then got a \"Submission Scoring Error\". To solve this I first added a cell copying sample_submission.csv as submision.csv into the working directory --> no error, so I knew that the error was not caused by my code and that the system also accepts incomplete submissions without error.\nThen I did the same for a submission made by my code on only the 10 public images --> error. So I knew that the error was caused by the formating of the file containing only the 10 public images. After schecking everything I could think of (double spaces, etc.), I went to https://regex101.com/ copied in the whole submission file and used the regex r\"0.\\d+ \\d+ \\d+ \\d+ \\d+\" to match all of my output. I saw that a few little parts did not match because the submission did not contain all 4 coordinates, only 3 in some rare cases. I then corrected the problem and it works now.\nHope this helps!\nEdit: I saw that you already fixed it. But maybe someone else will find this.
", "votes": null, "replies": [ { "id": 937330, "author_name": "chrisstan", "author_url": "", "post_date": "07/20/2020 23:56:42", "content": "Now it work! Thanks !!!!!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "934506": "I encounter submission error with this notebook.\nI am very confused because I have the right predicted bounding boxes there.\nBelow is my notebook:\nhttps://www.kaggle.com/chrisstan/yolov4-inference?scriptVersionId=39017659", "935492": "Not a solution but I was looking into your notebook and noticed although you have set your YoloV4 weights dataset private you copy those in the output folder resulting in it becoming publicly visible/available.\n\nIt would be a shame if some free loaders took advantage of this mistake. Please look into this issue and maybe setup a new revised notebook.", "935990": "Thank for your suggestion.\nI have fixed it.", "936270": "Your code looks similar to this so follow below solution-\nhttps://www.kaggle.com/c/global-wheat-detection/discussion/168179", "936780": "I tried like this and still failed..\n```python\nresults = []\n\nfor images, image_ids in data_loader:\n predictions = make_tta_predictions(images)\n try:\n for i, image in enumerate(images):\n\n boxes, scores, labels = run_wbf(predictions, image_index=i)\n boxes = (boxes*(1024 / 704)).round().astype(np.int32).clip(min=0, max=1023)\n image_id = image_ids[i]\n\n boxes[:, 2] = boxes[:, 2] - boxes[:, 0]\n boxes[:, 3] = boxes[:, 3] - boxes[:, 1]\n\n result = {\n 'image_id': image_id,\n 'PredictionString': format_prediction_string(boxes, scores)\n }\n results.append(result)\n except:\n result = {\n 'image_id': image_id,\n 'PredictionString': ''\n }\n results.append(result)\n```", "936856": "Check box length also like below-\n`if len(boxes)>0:\n result = {\n 'image_id': images_info['filename'][:-4],\n 'PredictionString': format_prediction_string(boxes, scores)}\n else:\n result = {\n 'image_id': images_info['filename'][:-4],\n 'PredictionString': ''}\n results.append(result)`\n\nAnd also put your `predictions = make_tta_predictions(images)` inside the try block.", "936864": "Just a wild guess but some non 704 sized image maybe causing an error in your `make_tta_predictions(images)`\nWe've too screwed a lot of our submissions over such rogue images which managed to reach the model input through some loophole in our code.\n\nTry copying all the test images (~1200 during submission) from the input folder to the output folder (kaggle/working/) and resize them all to your desired image size (704) right away.\nFeed this new test directory to your test loader function and see if the error still persists.\n\nOr just put a lazy condition like \n`if image.shape[0] != 704 or image.shape[1] != 704: image=cv2.resize(image, (704,704))\npredictions = make_tta_predictions(images)`\n\nYou can also locally test this theory of mine by taking some ~1000 train images as test images with weird shapes (take a few greater than 1024 some lesser than it and maybe try rectangle images etc.) and run your inference with them.\n\nSee if this helps.", "937200": "Hi, which submission error are you getting? I can only speak from my experience:\n1. First I got an \"Submission CSV not found\" error. This seems to have been caused by me having an additional folder in my working directory, so now I delete everything except submission.csv and the notebook. It could also have been caused by me copying too much data into the working directory, because I was copying the whole training set. I think that might have been too much (first competition errors).\n2. I then got a \"Submission Scoring Error\". To solve this I first added a cell copying sample_submission.csv as submision.csv into the working directory --> no error, so I knew that the error was not caused by my code and that the system also accepts incomplete submissions without error.\nThen I did the same for a submission made by my code on only the 10 public images --> error. So I knew that the error was caused by the formating of the file containing only the 10 public images. After schecking everything I could think of (double spaces, etc.), I went to https://regex101.com/ copied in the whole submission file and used the regex r\"0.\\d+ \\d+ \\d+ \\d+ \\d+\" to match all of my output. I saw that a few little parts did not match because the submission did not contain all 4 coordinates, only 3 in some rare cases. I then corrected the problem and it works now.\nHope this helps!\nEdit: I saw that you already fixed it. But maybe someone else will find this.", "937320": "Now it can submit! Thanks!", "937321": "Now it can work! Thanks!!!", "937330": "Now it work! Thanks !!!!!" }, "source": "meta" }