diff --git a/index.html b/index.html index fd49c9a..4944a90 100644 --- a/index.html +++ b/index.html @@ -159,6 +159,203 @@

Abstract

+
+
+
+
+

ICText Dataset

+

Samples of annotations in ICText dataset. Besides the common annotations + (i.e. bounding box and class), we also include the multi-label character quality attributes. + They are represented by (i) red dot for low contrast character, + (ii) green dot dot for blurry + characters, and lastly (iii) blue dot for broken characters.

+
+
+ ICText Sample 1 +
+ +
+ ICText Sample 2 +
+ +
+ ICText Sample 3 +
+
+

The ICText dataset contains a total of 10,000 images with 100,152 legible characters. + Of the 69,750 annotated samples in the training set, 50,597 have quality defects (72.5%). In the testing + set, 20,114 out of 30,402 characters (66.2%) have quality attributes. This + indicates a high number of flawed characters in ICText, which is also the + first-ever dataset with quality labels on characters. All + images and annotations are fairly distributed with a 7:3 ratio between the + training set (69.64% of complete annotations) and the testing set (30.36%). + Statistics of ICText are visualized in the following figures:

+ + +

The ICText dataset is also extended to a long-tail classification dataset, ICText-LT. + A new Frequency Weighted Focusing and Dynamic Probability Smoothing (FFDS) loss is then proposed to dynamically + reduces the influence of outliers and assigns class-dependent focusing parameters. + More details can be found in this GitHub repo.

+ +
+
+
+
+
+ +
+
+
+
+
+

AGCL

+

Attribute-Guided Curriculum Learning (AGCL) loss proposes to zero out the gradient of difficult characters. + It can also balances the contribution of negative samples through weighting factors and focusing parameters. + The training is split into two phases where AGCL loss is used in the first phase, and the Cross Entropy loss is used in the + second phase. +

+
+
+
+ +
Schematic diagram of AGCL.
+
+
+
+

The figure below shows the differences between common existing loss functions used by object detectors + and our proposed AGCL on positive and negative cases during training. +

+
+
+
+ +
Comparison of existing loss functions and AGCL.
+
+
+
+
+
+
+
+
+ +
+
+
+
+
+

Experiment Results

+

Quantitative results of all methods on ICText's test set are shown in the table below. + Models marked with * are tested on a subset of easier images. Both inference speed and GPU memory are tested on Titan + X, and the rest of the hardware specifications can be found in our paper. ± marks the + standard deviation calculated over five runs, and ↑ shows the relative AP improvement of + AGCL-enabled methods over the baseline methods.

+
+
+
+ +
Quantitative result table.
+
+
+
+

Qualitative results of ABCNet, PAN++, YOLOv4 (baseline), and our proposed + YOLOv4-AGCL are shown in the figures below. Green boxes = true positives; + Red boxes = false positives; + Blue boxes = false negatives. + The character class prediction is shown in the top left corner of each box.

+
+
+
+ +
(a) ABCNet.
+
+
+ +
+
+ +
(b) PAN++.
+
+
+
+
+
+
+ +
(c) YOLOv4 (baseline).
+
+
+ +
+
+ +
(d) YOLOv4-AGCL (Ours).
+
+
+
+ +

We show that asking non-AGCL detectors to learn directly from flawed characters has side effects, + i.e., more false positives and false negatives. Additionally, both ABCNet and PAN++ suffer from + granularity issues. In contrast, there are significantly fewer false-positive and false-negative boxes + in our proposed method (i.e., YOLOv4-AGCL), showing that training a detector in an + easy-to-hard fashion guided by quality attributes can achieve better results.

+
+
+
+
+
+ +
@@ -183,6 +380,17 @@

BibTeX

+

If you wish to cite the lastest version of the ICText dataset and AGCL:

+
+

+ + + Our paper is currently under review. We will update this section when it is published. + +

+
diff --git a/static/css/index.css b/static/css/index.css index 06b64a9..5c1f242 100644 --- a/static/css/index.css +++ b/static/css/index.css @@ -42,4 +42,14 @@ body { .tab { display: inline-block; margin-left: 15px; +} + +.resize { + object-fit: cover; + width: auto; + min-height: 280px; +} + +.figure-caption { + font-size: 1em; } \ No newline at end of file diff --git a/static/images/388715.jpg b/static/images/388715.jpg new file mode 100644 index 0000000..c3f83cf Binary files /dev/null and b/static/images/388715.jpg differ diff --git a/static/images/389836_abc.png b/static/images/389836_abc.png new file mode 100644 index 0000000..635c01f Binary files /dev/null and b/static/images/389836_abc.png differ diff --git a/static/images/389836_agcl.png b/static/images/389836_agcl.png new file mode 100644 index 0000000..3824d22 Binary files /dev/null and b/static/images/389836_agcl.png differ diff --git a/static/images/389836_baseline.png b/static/images/389836_baseline.png new file mode 100644 index 0000000..b32eda6 Binary files /dev/null and b/static/images/389836_baseline.png differ diff --git a/static/images/389836_panpp.png b/static/images/389836_panpp.png new file mode 100644 index 0000000..c662c93 Binary files /dev/null and b/static/images/389836_panpp.png differ diff --git a/static/images/59163.jpg b/static/images/59163.jpg new file mode 100644 index 0000000..abf08b4 Binary files /dev/null and b/static/images/59163.jpg differ diff --git a/static/images/91068.jpg b/static/images/91068.jpg new file mode 100644 index 0000000..8a8cd56 Binary files /dev/null and b/static/images/91068.jpg differ diff --git a/static/images/aes_stats.png b/static/images/aes_stats.png new file mode 100644 index 0000000..b89bf96 Binary files /dev/null and b/static/images/aes_stats.png differ diff --git a/static/images/agcl_stage.png b/static/images/agcl_stage.png new file mode 100644 index 0000000..29dc44b Binary files /dev/null and b/static/images/agcl_stage.png differ diff --git a/static/images/ann_stats.png b/static/images/ann_stats.png new file mode 100644 index 0000000..bf3324d Binary files /dev/null and b/static/images/ann_stats.png differ diff --git a/static/images/char_stats.png b/static/images/char_stats.png new file mode 100644 index 0000000..7b3aa8e Binary files /dev/null and b/static/images/char_stats.png differ diff --git a/static/images/loss_compare.png b/static/images/loss_compare.png new file mode 100644 index 0000000..464a5cd Binary files /dev/null and b/static/images/loss_compare.png differ diff --git a/static/images/quanti_res.png b/static/images/quanti_res.png new file mode 100644 index 0000000..a7af919 Binary files /dev/null and b/static/images/quanti_res.png differ