Merge remote-tracking branch 'origin/master'

coco/convert_to_pascalformat.py

@@ -0,0 +1,56 @@
+#!/usr/bin/python -tt
+
+# python code to convert coco val2014 JSON file to PASCAL XML.
+
+import sys
+
+def main():
+	if len(sys.argv) != 3:
+		print 'usage: python convert_to_pascalformat.py coco_dataDir coco_dataType'
+		print 'for example: python convert_to_pascalformat.py \'./\' \'val2014\''
+		sys.exit(1)
+
+	dataDir = sys.argv[1]
+	dataType = sys.argv[2]
+
+	from pycocotools.coco import COCO
+	import os
+
+	annFile='%s/annotations/instances_%s.json'%(dataDir,dataType)
+
+	coco=COCO(annFile)
+	cats = coco.loadCats(coco.getCatIds())
+	nms=[cat['name'] for cat in cats]
+
+	imgIds = coco.getImgIds()
+
+	directory = './annotations_pascalformat/'
+	if not os.path.exists(directory):
+	    os.makedirs(directory)
+
+	for n in xrange(len(imgIds)):
+		img = coco.loadImgs(imgIds[n])[0]
+		annIds = coco.getAnnIds(imgIds=img['id'], iscrowd=None)
+		anns = coco.loadAnns(annIds)
+
+		xml = '<annotation>\n<folder>\nCOCO2014pascalformat\n</folder>\n<filename>\n'
+		xml += img['file_name'] + '\n</filename>\n<source>\n<database>\nCOCO2014pascalformat\n</database>\n</source>\n<size>\n'
+		xml += '<width>\n' + str(img['width']) + '\n</width>\n' + '<height>\n' + str(img['height']) + '\n</height>\n'
+		xml += '<depth>\n3\n</depth>\n</size>\n<segmented>\n0\n</segmented>\n'
+
+		for i in xrange(len(anns)):
+			bbox = anns[i]['bbox']
+			xml += '<object>\n<name>\n' + str(anns[i]['category_id']) + '\n</name>\n'
+			xml += '<bndbox>\n<xmin>\n' + str(int(round(bbox[0]))) + '\n</xmin>\n'
+			xml += '<ymin>\n' + str(int(round(bbox[1]))) + '\n</ymin>\n'
+			xml += '<xmax>\n' + str(int(round(bbox[0] + bbox[2]))) + '\n</xmax>\n'
+			xml += '<ymax>\n' + str(int(round(bbox[1] + bbox[3]))) + '\n</ymax>\n</bndbox>\n'
+			xml += '<truncated>\n0\n</truncated>\n<difficult>\n0\n</difficult>\n</object>\n'
+		xml += '</annotation>'
+		f_xml = open(directory + img['file_name'].split('.jpg')[0] + '.xml', 'w')
+		f_xml.write(xml)
+		f_xml.close()
+		print str(n) + ' out of ' + str(len(imgIds))
+
+if __name__ == '__main__':
+  main()

test/coco_dpm_test.m

@@ -0,0 +1,138 @@
+function [ output ] = coco_dpm_test( dpm_model, dpm_thresh, dataDir, dataType, imgIds, outputFile )
+% coco_dpm_test is a wrapper of the DPM code running a trained dpm model 
+% on a set of images specified by their COCO ids.
+% Detections are:  
+%  - saved in the following JSON format: 
+% [{'image_id': int, 'cateogry_id': int, 'bbox': [x, y, w, h], 'score': float}]
+%  - returned in an output cell array
+%
+% USAGE:
+%  $ > load VOC2010/bicycle_final.mat
+%  $ > output = coco_dpm_test( model, -0.5, '../coco_dir', 'val2014', [388258], './bike_dpm_detections_coco.json' )
+%
+% INPUTS:
+%  - dpm_model      - model trained with dpm  
+%  - dpm_thresh     - threshold for detection of the trained model
+%  - dataDir        - path to the coco main directory
+%  - dataType       - train, test or validation type of data
+%  - imgIds         - array containing the COCO Ids of the images to use
+%  - outputFile     - path to the output file
+%
+% OUTPUTS:
+%  - output         - struct array containing an element for each detection
+%
+% Code written by Matteo Ruggero Ronchi, 2015.
+% Licensed under the Simplified BSD License [see coco/license.txt]
+
+output = struct();
+
+%% set dpm model parameters
+if nargin < 2
+    % if the threshold is not included we use the one in the model
+    dpm_thresh = dpm_model.thresh;
+end
+
+%% initialize COCO api for instance annotations
+if nargin < 3
+    dataDir = '../';
+end
+
+if nargin < 4
+    dataType = 'test2014';
+end
+
+annFile=sprintf('%s/annotations/instances_%s.json',dataDir,dataType);
+if(~exist('coco','var')), coco=CocoApi(annFile); end
+
+%% recover Ids of images to use for testing
+if nargin < 5
+    % if a list of IDs is not specified test on all the images
+    imgIds = coco.getImgIds();
+end
+nImages = length( imgIds );
+
+%% set maximum detection variables
+% maximum of 100 detections per image are allowed
+maxDetections = 100;
+counter = 0;
+
+%% adjuste difference in category names between COCO and PASCAL 
+switch dpm_model.class
+    case 'aeroplane'
+        dpm_class = 'airplane';
+    case 'diningtable'
+        dpm_class = 'dining table';
+    case 'motorbike'
+        dpm_class = 'motorcycle';
+    case 'pottedplant'
+        dpm_class = 'potted plant';
+    case 'sofa'
+        dpm_class = 'couch';
+    case 'tvmonitor'
+        dpm_class = 'tv';
+    otherwise
+        dpm_class = dpm_model.class;
+end     
+category_id = coco.getCatIds('catNms',dpm_class);
+
+for jj = 1:nImages
+
+    %% load and display image
+    try
+        img_info = coco.loadImgs(imgIds(jj));
+        img = imread(sprintf('%s/images/%s/%s',dataDir,dataType,img_info.file_name));
+
+        %figure(1); imagesc(img); axis('image'); set(gca,'XTick',[],'YTick',[])
+    catch
+        error('Error in imgIds vector. Id [%d] not in COCO! Stopping Test.', imgIds(jj) );
+    end
+
+    %% run dpm code on the image
+    [ds, bs] = imgdetect(img, dpm_model, dpm_thresh);
+
+    if ~isempty(ds)
+        if dpm_model.type == model_types.MixStar
+            if isfield(dpm_model, 'bboxpred')
+                bboxpred = dpm_model.bboxpred;
+                [ds, bs] = clipboxes(img, ds, bs);
+                [ds, bs] = bboxpred_get(bboxpred, ds, reduceboxes(dpm_model, bs));
+            else
+                warning('no bounding box predictor found');
+            end
+        end
+        [ds, bs] = clipboxes(img, ds, bs);
+        I = nms(ds, 0.5);
+        ds = ds(I,:);
+        bs = bs(I,:);
+    end
+
+    %% put the detections in COCO format
+
+    nDetections = size(ds,1);
+    if (nDetections > maxDetections)
+        warning('Number of detections in image [%s/%d] exceeds maximum allowed (%d). Detections in excess will not be stored.', dataType, imgIds(jj), maxDetections);
+        nDetections = maxDetections;
+    end
+    for ii = 1:nDetections
+        counter = counter + 1;
+
+        output(counter).image_id = imgIds(jj);
+        output(counter).category_id = category_id;
+        output(counter).bbox = ds(ii,1:4);
+        output(counter).score = ds(ii,5);
+    end
+
+end
+
+%% save detections on filesystem in JSON format and return output structure
+if nargin < 6
+    outputFile = strcat('./',dpm_class(~isspace(dpm_class)),'_dpm_detections_coco.json');
+end
+
+s = gason( output );
+
+fid = fopen(outputFile, 'w');
+fprintf(fid, '%s', s);
+fclose(fid);
+
+