init commit

.gitignore

@@ -99,3 +99,5 @@ ENV/
 
 # mypy
 .mypy_cache/
+
+*.pth

README.md

@@ -1,2 +1,19 @@
 # SFD_pytorch
 A PyTorch Implementation of Single Shot Scale-invariant Face Detector.
+
+# Eval
+```
+python wider_eval_pytorch.py
+
+cd eval/eval_tools_old-version
+octave wider_eval_pytorch.m
+```
+
+# Test
+```
+python test.py --model data/s3fd_convert.pth --path data/test01.jpg
+```
+![output](data/test01_output.png)
+
+# References
+[SFD](https://github.com/sfzhang15/SFD)

bbox.py

@@ -0,0 +1,95 @@
+from __future__ import print_function
+import os,sys,cv2,random,datetime,time,math
+import argparse
+import numpy as np
+import torch
+
+try:
+    from iou import IOU
+except:
+    # IOU cython speedup 10x
+    def IOU(ax1,ay1,ax2,ay2,bx1,by1,bx2,by2):
+        sa = abs((ax2-ax1)*(ay2-ay1))
+        sb = abs((bx2-bx1)*(by2-by1))
+        x1,y1 = max(ax1,bx1),max(ay1,by1)
+        x2,y2 = min(ax2,bx2),min(ay2,by2)
+        w  = x2 - x1
+        h  = y2 - y1
+        if w<0 or h<0: return 0.0
+        else: return 1.0*w*h/(sa+sb-w*h)
+
+def bboxlog(x1,y1,x2,y2,axc,ayc,aww,ahh):
+    xc,yc,ww,hh = (x2+x1)/2,(y2+y1)/2,x2-x1,y2-y1
+    dx,dy = (xc-axc)/aww,(yc-ayc)/ahh
+    dw,dh = math.log(ww/aww),math.log(hh/ahh)
+    return dx,dy,dw,dh
+
+def bboxloginv(dx,dy,dw,dh,axc,ayc,aww,ahh):
+    xc,yc = dx*aww+axc, dy*ahh+ayc
+    ww,hh = math.exp(dw)*aww,math.exp(dh)*ahh
+    x1,x2,y1,y2 = xc-ww/2,xc+ww/2,yc-hh/2,yc+hh/2
+    return x1,y1,x2,y2
+
+def nms(dets, thresh):
+    if 0==len(dets): return []
+    x1,y1,x2,y2,scores = dets[:, 0],dets[:, 1],dets[:, 2],dets[:, 3],dets[:, 4]
+    areas = (x2 - x1 + 1) * (y2 - y1 + 1)
+    order = scores.argsort()[::-1]
+
+    keep = []
+    while order.size > 0:
+        i = order[0]
+        keep.append(i)
+        xx1,yy1 = np.maximum(x1[i], x1[order[1:]]),np.maximum(y1[i], y1[order[1:]])
+        xx2,yy2 = np.minimum(x2[i], x2[order[1:]]),np.minimum(y2[i], y2[order[1:]])
+
+        w,h = np.maximum(0.0, xx2 - xx1 + 1),np.maximum(0.0, yy2 - yy1 + 1)
+        ovr = w*h / (areas[i] + areas[order[1:]] - w*h)
+
+        inds = np.where(ovr <= thresh)[0]
+        order = order[inds + 1]
+
+    return keep
+
+def encode(matched, priors, variances):
+    """Encode the variances from the priorbox layers into the ground truth boxes
+    we have matched (based on jaccard overlap) with the prior boxes.
+    Args:
+        matched: (tensor) Coords of ground truth for each prior in point-form
+            Shape: [num_priors, 4].
+        priors: (tensor) Prior boxes in center-offset form
+            Shape: [num_priors,4].
+        variances: (list[float]) Variances of priorboxes
+    Return:
+        encoded boxes (tensor), Shape: [num_priors, 4]
+    """
+
+    # dist b/t match center and prior's center
+    g_cxcy = (matched[:, :2] + matched[:, 2:])/2 - priors[:, :2]
+    # encode variance
+    g_cxcy /= (variances[0] * priors[:, 2:])
+    # match wh / prior wh
+    g_wh = (matched[:, 2:] - matched[:, :2]) / priors[:, 2:]
+    g_wh = torch.log(g_wh) / variances[1]
+    # return target for smooth_l1_loss
+    return torch.cat([g_cxcy, g_wh], 1)  # [num_priors,4]
+
+def decode(loc, priors, variances):
+    """Decode locations from predictions using priors to undo
+    the encoding we did for offset regression at train time.
+    Args:
+        loc (tensor): location predictions for loc layers,
+            Shape: [num_priors,4]
+        priors (tensor): Prior boxes in center-offset form.
+            Shape: [num_priors,4].
+        variances: (list[float]) Variances of priorboxes
+    Return:
+        decoded bounding box predictions
+    """
+
+    boxes = torch.cat((
+        priors[:, :2] + loc[:, :2] * variances[0] * priors[:, 2:],
+        priors[:, 2:] * torch.exp(loc[:, 2:] * variances[1])), 1)
+    boxes[:, :2] -= boxes[:, 2:] / 2
+    boxes[:, 2:] += boxes[:, :2]
+    return boxes

eval/eval_tools_old-version/boxoverlap.m

@@ -0,0 +1,35 @@
+function o = boxoverlap(a, b)
+% Compute the symmetric intersection over union overlap between a set of
+% bounding boxes in a and a single bounding box in b.
+%
+% a  a matrix where each row specifies a bounding box
+% b  a single bounding box
+
+% AUTORIGHTS
+% -------------------------------------------------------
+% Copyright (C) 2011-2012 Ross Girshick
+% Copyright (C) 2008, 2009, 2010 Pedro Felzenszwalb, Ross Girshick
+% 
+% This file is part of the voc-releaseX code
+% (http://people.cs.uchicago.edu/~rbg/latent/)
+% and is available under the terms of an MIT-like license
+% provided in COPYING. Please retain this notice and
+% COPYING if you use this file (or a portion of it) in
+% your project.
+% -------------------------------------------------------
+
+x1 = max(a(:,1), b(1));
+y1 = max(a(:,2), b(2));
+x2 = min(a(:,3), b(3));
+y2 = min(a(:,4), b(4));
+
+w = x2-x1+1;
+h = y2-y1+1;
+inter = w.*h;
+aarea = (a(:,3)-a(:,1)+1) .* (a(:,4)-a(:,2)+1);
+barea = (b(3)-b(1)+1) * (b(4)-b(2)+1);
+% intersection over union overlap
+o = inter ./ (aarea+barea-inter);
+% set invalid entries to 0 overlap
+o(w <= 0) = 0;
+o(h <= 0) = 0;

eval/eval_tools_old-version/evaluation.m

@@ -0,0 +1,99 @@
+function evaluation(norm_pred_list,gt_dir,setting_name,setting_class,legend_name)
+load(gt_dir);
+if ~exist(sprintf('./plot/baselines/Val/%s/%s',setting_class,legend_name),'dir')
+    mkdir(sprintf('./plot/baselines/Val/%s/%s',setting_class,legend_name));
+end
+IoU_thresh = 0.5;
+event_num = 61;
+thresh_num = 1000;
+org_pr_cruve = zeros(thresh_num,2);
+count_face = 0;
+
+for i = 1:event_num
+    img_list = file_list{i};
+    gt_bbx_list = face_bbx_list{i};
+    pred_list = norm_pred_list{i};
+    sub_gt_list = gt_list{i};
+    img_pr_info_list = cell(length(img_list),1);
+
+    fprintf('%s, current event %d\n',setting_name,i);
+    for j = 1:length(img_list)
+        gt_bbx = gt_bbx_list{j};
+        pred_info = pred_list{j};
+        keep_index = sub_gt_list{j};
+        count_face = count_face + length(keep_index);
+
+        if isempty(gt_bbx) || isempty(pred_info)
+            continue;
+        end
+        ignore = zeros(size(gt_bbx,1),1);
+        if ~isempty(keep_index)
+            ignore(keep_index) = 1;
+        end
+
+        [pred_recall, proposal_list] = image_evaluation(pred_info, gt_bbx, ignore, IoU_thresh);
+
+        img_pr_info = image_pr_info(thresh_num, pred_info, proposal_list, pred_recall);
+        img_pr_info_list{j} = img_pr_info;   
+    end
+    for j = 1:length(img_list)
+        img_pr_info = img_pr_info_list{j};
+        if ~isempty(img_pr_info)
+            org_pr_cruve(:,1) = org_pr_cruve(:,1) + img_pr_info(:,1);
+            org_pr_cruve(:,2) = org_pr_cruve(:,2) + img_pr_info(:,2);
+        end
+    end
+end
+pr_cruve = dataset_pr_info(thresh_num, org_pr_cruve, count_face);
+% save(sprintf('./plot/baselines/Val/%s/%s/wider_pr_info_%s_%s.mat',setting_class,legend_name,legend_name,setting_name),'pr_cruve','legend_name','-v7.3');
+save(sprintf('./plot/baselines/Val/%s/%s/wider_pr_info_%s_%s.mat',setting_class,legend_name,legend_name,setting_name),'pr_cruve','legend_name','-v7');
+end
+
+function [pred_recall,proposal_list] = image_evaluation(pred_info, gt_bbx, ignore, IoU_thresh)
+    pred_recall = zeros(size(pred_info,1),1);
+    recall_list = zeros(size(gt_bbx,1),1);
+    proposal_list = zeros(size(pred_info,1),1);
+    proposal_list = proposal_list + 1;
+    pred_info(:,3) = pred_info(:,1) + pred_info(:,3); 
+    pred_info(:,4) = pred_info(:,2) + pred_info(:,4);
+    gt_bbx(:,3) = gt_bbx(:,1) + gt_bbx(:,3);
+    gt_bbx(:,4) = gt_bbx(:,2) + gt_bbx(:,4);
+    for h = 1:size(pred_info,1)
+        overlap_list = boxoverlap(gt_bbx, pred_info(h,1:4));
+        [max_overlap, idx] = max(overlap_list);
+        if max_overlap >= IoU_thresh
+            if (ignore(idx) == 0)
+                recall_list(idx) = -1;
+                proposal_list(h) = -1;
+            elseif (recall_list(idx)==0)
+                recall_list(idx) = 1;
+            end
+        end
+        r_keep_index = find(recall_list == 1);
+        pred_recall(h) = length(r_keep_index);
+    end
+end
+
+function img_pr_info = image_pr_info(thresh_num, pred_info, proposal_list, pred_recall)
+    img_pr_info = zeros(thresh_num,2);       
+    for t = 1:thresh_num
+        thresh = 1-t/thresh_num;
+        r_index = find(pred_info(:,5)>=thresh,1,'last');
+        if (isempty(r_index))
+            img_pr_info(t,2) = 0;
+            img_pr_info(t,1) = 0;
+        else
+            p_index = find(proposal_list(1:r_index) == 1);
+            img_pr_info(t,1) = length(p_index);
+            img_pr_info(t,2) = pred_recall(r_index);
+        end
+    end
+end
+
+function pr_cruve = dataset_pr_info(thresh_num, org_pr_cruve, count_face)
+    pr_cruve = zeros(thresh_num,2);
+    for i = 1:thresh_num
+        pr_cruve(i,1) = org_pr_cruve(i,2)/org_pr_cruve(i,1);
+        pr_cruve(i,2) = org_pr_cruve(i,2)/count_face;
+    end
+end

eval/eval_tools_old-version/nms.m

@@ -0,0 +1,56 @@
+function pick = nms(boxes, overlap)
+% top = nms(boxes, overlap)
+% Non-maximum suppression. (FAST VERSION)
+% Greedily select high-scoring detections and skip detections
+% that are significantly covered by a previously selected
+% detection.
+%
+% NOTE: This is adapted from Pedro Felzenszwalb's version (nms.m),
+% but an inner loop has been eliminated to significantly speed it
+% up in the case of a large number of boxes
+
+% Copyright (C) 2011-12 by Tomasz Malisiewicz
+% All rights reserved.
+% 
+% This file is part of the Exemplar-SVM library and is made
+% available under the terms of the MIT license (see COPYING file).
+% Project homepage: https://github.com/quantombone/exemplarsvm
+
+
+if isempty(boxes)
+  pick = [];
+  return;
+end
+
+x1 = boxes(:,1);
+y1 = boxes(:,2);
+x2 = boxes(:,3);
+y2 = boxes(:,4);
+s = boxes(:,end);
+
+area = (x2-x1+1) .* (y2-y1+1);
+[vals, I] = sort(s);
+
+pick = s*0;
+counter = 1;
+while ~isempty(I)
+  last = length(I);
+  i = I(last);  
+  pick(counter) = i;
+  counter = counter + 1;
+
+  xx1 = max(x1(i), x1(I(1:last-1)));
+  yy1 = max(y1(i), y1(I(1:last-1)));
+  xx2 = min(x2(i), x2(I(1:last-1)));
+  yy2 = min(y2(i), y2(I(1:last-1)));
+
+  w = max(0.0, xx2-xx1+1);
+  h = max(0.0, yy2-yy1+1);
+
+  inter = w.*h;
+  o = inter ./ (area(i) + area(I(1:last-1)) - inter);
+
+  I = I(find(o<=overlap));
+end
+
+pick = pick(1:(counter-1));

eval/eval_tools_old-version/norm_score.m

@@ -0,0 +1,31 @@
+function norm_pred_list = norm_score(org_pred_list)
+
+event_num = 61;
+norm_pred_list = cell(event_num,1);
+max_score = realmin('single');
+min_score = realmax('single');
+parfor i = 1:event_num
+    pred_list = org_pred_list{i};
+    for j = 1:size(pred_list,1)
+        if(isempty(pred_list{j}))
+            continue;
+        end
+        score_list = pred_list{j}(:,5);
+        max_score = max(max_score,max(score_list));
+        min_score = min(min_score,min(score_list));
+    end
+end
+
+parfor i = 1:event_num
+    fprintf('Norm prediction: current event %d\n',i);
+    pred_list = org_pred_list{i};
+    for j = 1:size(pred_list,1)
+        if(isempty(pred_list{j}))
+            continue;
+        end
+        score_list = pred_list{j}(:,5);
+        norm_score_list = (score_list - min_score)/(max_score - min_score);
+        pred_list{j}(:,5) = norm_score_list;
+    end
+    norm_pred_list{i} = pred_list;
+end

eval/eval_tools_old-version/plot/VOCap.m

@@ -0,0 +1,9 @@
+function ap = VOCap(rec,prec)
+
+mrec=[0 ; rec ; 1];
+mpre=[0 ; prec ; 0];
+for i=numel(mpre)-1:-1:1
+    mpre(i)=max(mpre(i),mpre(i+1));
+end
+i=find(mrec(2:end)~=mrec(1:end-1))+1;
+ap=sum((mrec(i)-mrec(i-1)).*mpre(i));

eval/eval_tools_old-version/plot/plot_pr.m

@@ -0,0 +1,37 @@
+function plot_pr(propose,recall,lendge_name,seting_class,setting_name,dateset_class)
+model_num = size(propose,1);
+figure1 = figure('PaperSize',[20.98 29.68],'Color',[1 1 1]);
+axes1 = axes('Parent',figure1,...
+    'LineWidth',2,...
+    'FontSize',15,...
+    'FontName','Times New Roman',...
+    'FontWeight','bold');
+box(axes1,'on');
+hold on;
+
+LineColor = colormap(hsv(model_num));
+for i=1:model_num
+    plot(propose{i},recall{i},...
+        'MarkerEdgeColor',LineColor(i,:),...
+        'MarkerFaceColor',LineColor(i,:),...
+        'LineWidth',4,...
+        'Color',LineColor(i,:))
+    hleg = legend(lendge_name{:},'Location','SouthEast');
+    grid on;
+    hold on;
+end
+xlim([0,1]);
+ylim([0,1]);
+xlabel('Recall');
+ylabel('Precision');
+
+% savename = sprintf('./plot/figure/%s/wider_pr_cruve_%s_%s.pdf',dateset_class,seting_class,setting_name);
+savename = sprintf('./plot/figure_%s_wider_pr_cruve_%s_%s.jpg',dateset_class,seting_class,setting_name);
+print("-djpg",savename)
+% saveTightFigure(gcf,savename);
+clear gcf;
+hold off;
+
+
+
+