From e25168f042f19df29aa371327a4f167646e625a4 Mon Sep 17 00:00:00 2001 From: Peter Anderson Date: Tue, 25 Jun 2019 12:05:24 -0400 Subject: [PATCH 1/5] Update links to trained models and added user-pays google cloud storage download links --- README.md | 21 +++++++++++++++++++-- 1 file changed, 19 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index 94c25c2..138507a 100755 --- a/README.md +++ b/README.md @@ -48,6 +48,23 @@ For ease-of-use, we make pretrained features available for the entire [MSCOCO da - [2014 Testing Image Features (40K / 9GB)](https://imagecaption.blob.core.windows.net/imagecaption/test2014_36.zip) - [2015 Testing Image Features (80K / 17GB)](https://imagecaption.blob.core.windows.net/imagecaption/test2015_36.zip) + +If the above links are too slow, the features can be downloaded from Google Cloud Storage at own cost using the Requester Pays feature. Use the following gsutil commands where [PROJECT_ID] is your Cloud Storage project id that should be billed (further documentation can be found [here](https://cloud.google.com/storage/docs/using-requester-pays): + +10 to 100 features per image (adaptive): 10 to 100 features per image (adaptive): +``` +gsutil -u [PROJECT_ID] cp gs://bottom-up-attention/trainval.zip [OBJECT_DESTINATION] # 2014 Train/Val Image Features (120K / 23GB) +gsutil -u [PROJECT_ID] cp gs://bottom-up-attention/test2014.zip [OBJECT_DESTINATION] # 2014 Testing Image Features (40K / 7.3GB) +gsutil -u [PROJECT_ID] cp gs://bottom-up-attention/test2015.zip [OBJECT_DESTINATION] # 2015 Testing Image Features (80K / 15GB) +``` + +36 features per image (fixed): 36 features per image (fixed): +``` +gsutil -u [PROJECT_ID] cp gs://bottom-up-attention/trainval_36.zip [OBJECT_DESTINATION] # 2014 Train/Val Image Features (120K / 25GB) +gsutil -u [PROJECT_ID] cp gs://bottom-up-attention/test2014_36.zip [OBJECT_DESTINATION] # 2014 Testing Image Features (40K / 9GB) +gsutil -u [PROJECT_ID] cp gs://bottom-up-attention/test2015_36.zip [OBJECT_DESTINATION] # 2015 Testing Image Features (80K / 17GB) +``` + Both sets of features can be recreated by using `tools/genenerate_tsv.py` with the appropriate pretrained model and with MIN_BOXES/MAX_BOXES set to either 10/100 or 36/36 respectively - refer [Demo](#demo). ### Contents @@ -105,11 +122,11 @@ Any NVIDIA GPU with 12GB or larger memory is OK for training Faster R-CNN ResNet ### Demo -1. Download [pretrained model](https://storage.googleapis.com/bottom-up-attention/resnet101_faster_rcnn_final.caffemodel), and put it under `data\faster_rcnn_models`. +1. Download [pretrained model](https://www.dropbox.com/s/tr24q7h0zm2wnjv/resnet101_faster_rcnn_final.caffemodel?dl=1), and put it under `data\faster_rcnn_models`. 2. Run `tools/demo.ipynb` to show object and attribute detections on demo images. -3. Run `tools/genenerate_tsv.py` to extract bounding box features to a tab-separated-values (tsv) file. This will require modifying the `load_image_ids` function to suit your data locations. To recreate the pretrained feature files with 10 to 100 features per image, set MIN_BOXES=10 and MAX_BOXES=100. To recreate the pretrained feature files with 36 features per image, set MIN_BOXES=36 and MAX_BOXES=36 use this [alternative pretrained model](https://storage.googleapis.com/bottom-up-attention/resnet101_faster_rcnn_final_iter_320000.caffemodel) instead. The alternative pretrained model was trained for fewer iterations but performance is similar. +3. Run `tools/genenerate_tsv.py` to extract bounding box features to a tab-separated-values (tsv) file. This will require modifying the `load_image_ids` function to suit your data locations. To recreate the pretrained feature files with 10 to 100 features per image, set MIN_BOXES=10 and MAX_BOXES=100. To recreate the pretrained feature files with 36 features per image, set MIN_BOXES=36 and MAX_BOXES=36 use this [alternative pretrained model](https://www.dropbox.com/s/bacig173qnxddvz/resnet101_faster_rcnn_final_iter_320000.caffemodel?dl=1) instead. The alternative pretrained model was trained for fewer iterations but performance is similar. ### Training From 151cea129f96aa9dd52c62810933236f89ec53e4 Mon Sep 17 00:00:00 2001 From: Peter Anderson Date: Tue, 25 Jun 2019 12:07:05 -0400 Subject: [PATCH 2/5] minor typo in readme --- README.md | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/README.md b/README.md index 138507a..a5df148 100755 --- a/README.md +++ b/README.md @@ -49,16 +49,16 @@ For ease-of-use, we make pretrained features available for the entire [MSCOCO da - [2015 Testing Image Features (80K / 17GB)](https://imagecaption.blob.core.windows.net/imagecaption/test2015_36.zip) -If the above links are too slow, the features can be downloaded from Google Cloud Storage at own cost using the Requester Pays feature. Use the following gsutil commands where [PROJECT_ID] is your Cloud Storage project id that should be billed (further documentation can be found [here](https://cloud.google.com/storage/docs/using-requester-pays): +If the above links are too slow, the features can be downloaded from Google Cloud Storage at own cost using the Requester Pays feature. Use the following gsutil commands where [PROJECT_ID] is your Cloud Storage project id that should be billed (further documentation can be found [here](https://cloud.google.com/storage/docs/using-requester-pays)): -10 to 100 features per image (adaptive): 10 to 100 features per image (adaptive): +10 to 100 features per image (adaptive): ``` gsutil -u [PROJECT_ID] cp gs://bottom-up-attention/trainval.zip [OBJECT_DESTINATION] # 2014 Train/Val Image Features (120K / 23GB) gsutil -u [PROJECT_ID] cp gs://bottom-up-attention/test2014.zip [OBJECT_DESTINATION] # 2014 Testing Image Features (40K / 7.3GB) gsutil -u [PROJECT_ID] cp gs://bottom-up-attention/test2015.zip [OBJECT_DESTINATION] # 2015 Testing Image Features (80K / 15GB) ``` -36 features per image (fixed): 36 features per image (fixed): +36 features per image (fixed): ``` gsutil -u [PROJECT_ID] cp gs://bottom-up-attention/trainval_36.zip [OBJECT_DESTINATION] # 2014 Train/Val Image Features (120K / 25GB) gsutil -u [PROJECT_ID] cp gs://bottom-up-attention/test2014_36.zip [OBJECT_DESTINATION] # 2014 Testing Image Features (40K / 9GB) From 4bb21fbd73cbdd58ca1cbfbb092ca841620f4079 Mon Sep 17 00:00:00 2001 From: Peter Anderson Date: Tue, 12 Nov 2019 08:57:59 -0500 Subject: [PATCH 3/5] Fix broken links to pretrained models --- README.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index a5df148..4a795e7 100755 --- a/README.md +++ b/README.md @@ -122,11 +122,11 @@ Any NVIDIA GPU with 12GB or larger memory is OK for training Faster R-CNN ResNet ### Demo -1. Download [pretrained model](https://www.dropbox.com/s/tr24q7h0zm2wnjv/resnet101_faster_rcnn_final.caffemodel?dl=1), and put it under `data\faster_rcnn_models`. +1. Download [pretrained model](https://www.dropbox.com/s/wqada4qiv1dz9dk/resnet101_faster_rcnn_final.caffemodel?dl=1), and put it under `data\faster_rcnn_models`. 2. Run `tools/demo.ipynb` to show object and attribute detections on demo images. -3. Run `tools/genenerate_tsv.py` to extract bounding box features to a tab-separated-values (tsv) file. This will require modifying the `load_image_ids` function to suit your data locations. To recreate the pretrained feature files with 10 to 100 features per image, set MIN_BOXES=10 and MAX_BOXES=100. To recreate the pretrained feature files with 36 features per image, set MIN_BOXES=36 and MAX_BOXES=36 use this [alternative pretrained model](https://www.dropbox.com/s/bacig173qnxddvz/resnet101_faster_rcnn_final_iter_320000.caffemodel?dl=1) instead. The alternative pretrained model was trained for fewer iterations but performance is similar. +3. Run `tools/genenerate_tsv.py` to extract bounding box features to a tab-separated-values (tsv) file. This will require modifying the `load_image_ids` function to suit your data locations. To recreate the pretrained feature files with 10 to 100 features per image, set MIN_BOXES=10 and MAX_BOXES=100. To recreate the pretrained feature files with 36 features per image, set MIN_BOXES=36 and MAX_BOXES=36 use this [alternative pretrained model](https://www.dropbox.com/s/nu6jwhc88ujbw1v/resnet101_faster_rcnn_final_iter_320000.caffemodel?dl=1) instead. The alternative pretrained model was trained for fewer iterations but performance is similar. ### Training From ec7422aa6b672ff8c8a3594805cbe269cbf29723 Mon Sep 17 00:00:00 2001 From: Peter Date: Sun, 5 Apr 2020 12:18:51 -0500 Subject: [PATCH 4/5] Update dropbox links --- README.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index 4a795e7..4a9ab29 100755 --- a/README.md +++ b/README.md @@ -122,11 +122,11 @@ Any NVIDIA GPU with 12GB or larger memory is OK for training Faster R-CNN ResNet ### Demo -1. Download [pretrained model](https://www.dropbox.com/s/wqada4qiv1dz9dk/resnet101_faster_rcnn_final.caffemodel?dl=1), and put it under `data\faster_rcnn_models`. +1. Download [pretrained model](https://www.dropbox.com/s/5xethd2nxa8qrnq/resnet101_faster_rcnn_final.caffemodel?dl=1), and put it under `data\faster_rcnn_models`. 2. Run `tools/demo.ipynb` to show object and attribute detections on demo images. -3. Run `tools/genenerate_tsv.py` to extract bounding box features to a tab-separated-values (tsv) file. This will require modifying the `load_image_ids` function to suit your data locations. To recreate the pretrained feature files with 10 to 100 features per image, set MIN_BOXES=10 and MAX_BOXES=100. To recreate the pretrained feature files with 36 features per image, set MIN_BOXES=36 and MAX_BOXES=36 use this [alternative pretrained model](https://www.dropbox.com/s/nu6jwhc88ujbw1v/resnet101_faster_rcnn_final_iter_320000.caffemodel?dl=1) instead. The alternative pretrained model was trained for fewer iterations but performance is similar. +3. Run `tools/genenerate_tsv.py` to extract bounding box features to a tab-separated-values (tsv) file. This will require modifying the `load_image_ids` function to suit your data locations. To recreate the pretrained feature files with 10 to 100 features per image, set MIN_BOXES=10 and MAX_BOXES=100. To recreate the pretrained feature files with 36 features per image, set MIN_BOXES=36 and MAX_BOXES=36 use this [alternative pretrained model](https://www.dropbox.com/s/2h4hmgcvpaewizu/resnet101_faster_rcnn_final_iter_320000.caffemodel?dl=1) instead. The alternative pretrained model was trained for fewer iterations but performance is similar. ### Training From 20113f02f4d77baa790ccd30630f395144fe66a0 Mon Sep 17 00:00:00 2001 From: Peter Anderson Date: Sun, 14 Feb 2021 10:57:18 -0600 Subject: [PATCH 5/5] Provides faster download links. Closes #96, closes #93, closes #84, closes #72, closes #69. --- README.md | 35 +++++++++-------------------------- 1 file changed, 9 insertions(+), 26 deletions(-) diff --git a/README.md b/README.md index 4a9ab29..eabaa2d 100755 --- a/README.md +++ b/README.md @@ -36,34 +36,17 @@ bottom-up-attention is released under the MIT License (refer to the LICENSE file For ease-of-use, we make pretrained features available for the entire [MSCOCO dataset](http://mscoco.org/dataset/#download). It is not necessary to clone or build this repo to use features downloaded from the links below. Features are stored in tsv (tab-separated-values) format that can be read with `tools/read_tsv.py`. -**LINKS HAVE BEEN UPDATED** +**LINKS HAVE BEEN UPDATED TO GOOGLE CLOUD STORAGE (14 Feb 2021)** 10 to 100 features per image (adaptive): -- [2014 Train/Val Image Features (120K / 23GB)](https://imagecaption.blob.core.windows.net/imagecaption/trainval.zip) -- [2014 Testing Image Features (40K / 7.3GB)](https://imagecaption.blob.core.windows.net/imagecaption/test2014.zip) -- [2015 Testing Image Features (80K / 15GB)](https://imagecaption.blob.core.windows.net/imagecaption/test2015.zip) +- [2014 Train/Val Image Features (120K / 23GB)](https://storage.googleapis.com/up-down-attention/trainval.zip) +- [2014 Testing Image Features (40K / 7.3GB)](https://storage.googleapis.com/up-down-attention/test2014.zip) +- [2015 Testing Image Features (80K / 15GB)](https://storage.googleapis.com/up-down-attention/test2015.zip) 36 features per image (fixed): -- [2014 Train/Val Image Features (120K / 25GB)](https://imagecaption.blob.core.windows.net/imagecaption/trainval_36.zip) -- [2014 Testing Image Features (40K / 9GB)](https://imagecaption.blob.core.windows.net/imagecaption/test2014_36.zip) -- [2015 Testing Image Features (80K / 17GB)](https://imagecaption.blob.core.windows.net/imagecaption/test2015_36.zip) - - -If the above links are too slow, the features can be downloaded from Google Cloud Storage at own cost using the Requester Pays feature. Use the following gsutil commands where [PROJECT_ID] is your Cloud Storage project id that should be billed (further documentation can be found [here](https://cloud.google.com/storage/docs/using-requester-pays)): - -10 to 100 features per image (adaptive): -``` -gsutil -u [PROJECT_ID] cp gs://bottom-up-attention/trainval.zip [OBJECT_DESTINATION] # 2014 Train/Val Image Features (120K / 23GB) -gsutil -u [PROJECT_ID] cp gs://bottom-up-attention/test2014.zip [OBJECT_DESTINATION] # 2014 Testing Image Features (40K / 7.3GB) -gsutil -u [PROJECT_ID] cp gs://bottom-up-attention/test2015.zip [OBJECT_DESTINATION] # 2015 Testing Image Features (80K / 15GB) -``` - -36 features per image (fixed): -``` -gsutil -u [PROJECT_ID] cp gs://bottom-up-attention/trainval_36.zip [OBJECT_DESTINATION] # 2014 Train/Val Image Features (120K / 25GB) -gsutil -u [PROJECT_ID] cp gs://bottom-up-attention/test2014_36.zip [OBJECT_DESTINATION] # 2014 Testing Image Features (40K / 9GB) -gsutil -u [PROJECT_ID] cp gs://bottom-up-attention/test2015_36.zip [OBJECT_DESTINATION] # 2015 Testing Image Features (80K / 17GB) -``` +- [2014 Train/Val Image Features (120K / 25GB)](https://storage.googleapis.com/up-down-attention/trainval_36.zip) +- [2014 Testing Image Features (40K / 9GB)](https://storage.googleapis.com/up-down-attention/test2014_36.zip) +- [2015 Testing Image Features (80K / 17GB)](https://storage.googleapis.com/up-down-attention/test2015_36.zip) Both sets of features can be recreated by using `tools/genenerate_tsv.py` with the appropriate pretrained model and with MIN_BOXES/MAX_BOXES set to either 10/100 or 36/36 respectively - refer [Demo](#demo). @@ -122,11 +105,11 @@ Any NVIDIA GPU with 12GB or larger memory is OK for training Faster R-CNN ResNet ### Demo -1. Download [pretrained model](https://www.dropbox.com/s/5xethd2nxa8qrnq/resnet101_faster_rcnn_final.caffemodel?dl=1), and put it under `data\faster_rcnn_models`. +1. Download [pretrained model](https://storage.googleapis.com/up-down-attention/resnet101_faster_rcnn_final.caffemodel), and put it under `data\faster_rcnn_models`. 2. Run `tools/demo.ipynb` to show object and attribute detections on demo images. -3. Run `tools/genenerate_tsv.py` to extract bounding box features to a tab-separated-values (tsv) file. This will require modifying the `load_image_ids` function to suit your data locations. To recreate the pretrained feature files with 10 to 100 features per image, set MIN_BOXES=10 and MAX_BOXES=100. To recreate the pretrained feature files with 36 features per image, set MIN_BOXES=36 and MAX_BOXES=36 use this [alternative pretrained model](https://www.dropbox.com/s/2h4hmgcvpaewizu/resnet101_faster_rcnn_final_iter_320000.caffemodel?dl=1) instead. The alternative pretrained model was trained for fewer iterations but performance is similar. +3. Run `tools/genenerate_tsv.py` to extract bounding box features to a tab-separated-values (tsv) file. This will require modifying the `load_image_ids` function to suit your data locations. To recreate the pretrained feature files with 10 to 100 features per image, set MIN_BOXES=10 and MAX_BOXES=100. To recreate the pretrained feature files with 36 features per image, set MIN_BOXES=36 and MAX_BOXES=36 use this [alternative pretrained model](https://storage.googleapis.com/up-down-attention/resnet101_faster_rcnn_final_iter_320000.caffemodel) instead. The alternative pretrained model was trained for fewer iterations but performance is similar. ### Training