Development

.github/workflows/tests.yaml

@@ -7,7 +7,7 @@ on:
   push:
     branches: [ "cleanup" ]
   pull_request:
-    branches: [ "main",  ]
+    branches: [ "main", "development" ]
 
 permissions:
   contents: read

CITATION.cff

@@ -10,8 +10,8 @@ authors:
   - name: Samuel J. Cooper
     orcid: 0000-0003-4055-6903
 title: "Prediction of Microstructural Representativity from a Single Image"
-doi: ARXIV_DOI
-url: "https://github.com/tldr-group/ImageRep"
+doi: arXiv:2410.19568v1 
+url: "https://arxiv.org/abs/2410.19568v1"
 preferred-citation:
   type: article
   authors:
@@ -23,7 +23,7 @@ preferred-citation:
     orcid: 0000-0002-7263-6724
   - name: Samuel J. Cooper
     orcid: 0000-0003-4055-6903
-  doi: ARXIV_DOI
+  doi: arXiv:2410.19568v1 
   journal: "arXiV preprint"
   month: 8
   title: "Prediction of Microstructural Representativity from a Single Image"

README.md

@@ -2,7 +2,7 @@
 
 [Try it out!](https://www.imagerep.io/)
 
-Here we introduce the 'ImageRep' method for fast phase fraction representativity estimation from a single microstructural image. This is achieved by calculating the Two-Point Correlation (TPC) function of the image, combined with a data-driven analysis of the [MicroLib](https://microlib.io/) dataset. By applying a statistical framework that utilizes both data sources, we can establish the uncertainty in the phase fraction in the image with a given confidence, **and** the image size that would be needed to meet a given target uncertainty. Further details are provided in our [paper](CITATION.cff).
+Here we introduce the 'ImageRep' method for fast phase fraction representativity estimation from a single microstructural image. This is achieved by calculating the Two-Point Correlation (TPC) function of the image, combined with a data-driven analysis of the [MicroLib](https://microlib.io/) dataset. By applying a statistical framework that utilizes both data sources, we can establish the uncertainty in the phase fraction in the image with a given confidence, **and** the image size that would be needed to meet a given target uncertainty. Further details are provided in our [preprint](https://arxiv.org/abs/2410.19568).
 
 If you use this ImageRep in your research, [please cite us](CITATION.cff).
 

frontend/src/components/DragDrop.tsx

@@ -44,9 +44,12 @@ const DragDrop = ({ loadFromFile }: DragDropProps): JSX.Element => {
             onDragOver={handleDrag}
             onDrop={handeDrop}
         >
-            {(!isMobile) && <span>Drag microstructure file, or view example <a style={{ cursor: "pointer", color: 'blue' }} onClick={e => viewExample(DEFAULT_IMAGE_2D)}>in 2D</a> or  <a style={{ cursor: "pointer", color: 'blue' }} onClick={e => viewExample(DEFAULT_IMAGE_3D)}> 3D</a></span>}
+            {(!isMobile) && <div style={{display: 'flex', alignItems: 'center', flexDirection: 'column'}}>
+                <span>Drag microstructure file, or view example <a style={{ cursor: "pointer", color: 'blue' }} onClick={e => viewExample(DEFAULT_IMAGE_2D)}>in 2D</a> or  <a style={{ cursor: "pointer", color: 'blue' }} onClick={e => viewExample(DEFAULT_IMAGE_3D)}> 3D</a></span>
+                <span style={{fontSize: '0.7em'}}>(image must be segmented & image background (e.g. scale bar) cropped out)</span>
+            </div>}
         </div>
     );
 }
 
-export default DragDrop
+export default DragDrop

frontend/src/components/Menu.tsx

@@ -98,14 +98,34 @@ const ConfidenceSelect = () => {
 
 
     const vals = imageInfo?.phaseVals!
+
+    console.log(accurateFractions)
+    console.log(vals)
+    console.log(selectedPhase)
+
+    const displayPhaseFrac = () => {
+        try {
+        if (accurateFractions == null && imageInfo && vals[selectedPhase -1] && imageInfo?.previewData.data) {
+            console.log('help')
+            const pf = getPhaseFraction(
+                imageInfo?.previewData.data!,
+                vals[selectedPhase - 1]
+            )
+            if (typeof pf == "number") {
+                return pf.toFixed(3)
+            }
+        } else if (accurateFractions && vals[selectedPhase -1]) {
+            return accurateFractions[vals[selectedPhase - 1]].toFixed(3)
+        }
+
+        return ''
+    } catch (e) {
+        return ''
+    }
+    }
     // horrible ternary: if server has responded and set the accurate phase fractions,
     // then use those values in the modal. If not, use the estimate from the first image
-    const phaseFrac = (accurateFractions != null) ?
-        accurateFractions[vals[selectedPhase - 1]].toFixed(3)
-        : getPhaseFraction(
-            imageInfo?.previewData.data!,
-            vals[selectedPhase - 1]
-        ).toFixed(3);
+    const phaseFrac = displayPhaseFrac()
 
     const setConf = (e: any) => {
         setSelectedConf(Number(e.target!.value))
@@ -236,13 +256,6 @@ const Result = () => {
         setMenuState('processing');
     }
 
-    const getDPofSigFig = (decimal: number) => {
-        const rounded = parseFloat(decimal.toPrecision(1));
-        const loc = Math.ceil(Math.abs(Math.log10(rounded)));
-        const capped = Math.min(loc, 5)
-        return capped
-    }
-
     const c = colours[selectedPhase];
     const headerHex = rgbaToHex(c[0], c[1], c[2], c[3]);
 
@@ -255,10 +268,8 @@ const Result = () => {
     const absErrFromPFB = (pfB![1] - pfB![0]) / 2
     const perErrFromPFB = 100 * (((pfB![1] - pfB![0]) / 2) / phaseFrac)
 
-    const roundTo = getDPofSigFig(absErrFromPFB);
-
     const beforeBoldText = `The phase fraction in the segmented image is ${phaseFrac.toFixed(3)}. Assuming perfect segmentation, the 'ImageRep' model proposed by Dahari et al. suggests that `
-    const boldText = `we can be ${selectedConf.toFixed(1)}% confident that the material's phase fraction is within ${perErrFromPFB?.toFixed(1)}% of this value (i.e. ${phaseFrac.toFixed(roundTo)}±${(absErrFromPFB).toFixed(roundTo)})`
+    const boldText = `we can be ${selectedConf.toFixed(1)}% confident that the material's phase fraction is within ${perErrFromPFB?.toFixed(1)}% of this value (i.e. ${phaseFrac.toFixed(3)}±${absErrFromPFB.toFixed(3)})`
     const copyText = beforeBoldText + boldText
     const afterText = "These results are derived from an estimated "
     const afterBoldText = `Characteristic Length Scale (CLS) of ${analysisInfo?.integralRange!.toFixed(0)}px`
@@ -376,7 +387,7 @@ const Result = () => {
                             <ListGroup>
                                 <ListGroup.Item variant="dark" style={{ cursor: "pointer" }} onClick={e => handleShowFull()}>Click for Brief explanation!</ListGroup.Item>
                                 <ListGroup.Item>Implementation in the <a href="https://github.com/tldr-group/Representativity">GitHub</a></ListGroup.Item>
-                                <ListGroup.Item>Full details can be found in the <a href="comingsoon">paper</a></ListGroup.Item>
+                                <ListGroup.Item>Full details can be found in the <a href="https://arxiv.org/abs/2410.19568v1">paper</a></ListGroup.Item>
                             </ListGroup>
                         </Accordion.Body>
                     </Accordion.Item>

frontend/src/components/Popups.tsx

@@ -128,7 +128,8 @@ export const MoreInfo = () => {
                             the measured phase fraction with c% of the time.</b>
                     </p>
 
-                    <p>Full details can be found in the <a href="comingsoon">paper</a>.</p>
+                    <p>Full details can be found in the <a href="https://arxiv.org/abs/2410.19568v1">paper</a>.</p>
+                    <p>Source code is available <a href="https://github.com/tldr-group/ImageRep">here</a>; please request any features or report any bugs in <a href="https://github.com/tldr-group/ImageRep/issues">the issues page!</a></p>
                 </Modal.Body>
                 <Modal.Footer>
                     <Button variant="dark" onClick={handleClose}>

frontend/src/components/imageLogic.tsx

@@ -35,9 +35,9 @@ export const replaceGreyscaleWithColours = (arr: Uint8ClampedArray, mapping: { [
 }
 
 export const getPhaseFraction = (arr: Uint8ClampedArray, val: number, nChannels: number = 4) => {
-    console.log('ahhhh')
     const uniqueVals = arr.filter((_, i, __) => { return i % nChannels == 0 })
     const matching = uniqueVals.filter((v) => v == val);
+    if (arr.length == 0) {return 0}
     return (matching.length) / (arr.length / nChannels);
 }
 

paper_figures/introduction.py

@@ -51,7 +51,7 @@
     sofc_small_im = sofc_large_im[y1:y2, x1:x2]
     ax_sofc_im = fig.add_subplot(gs[0, 0])
     ax_sofc_im.imshow(sofc_large_im, cmap='gray', interpolation='nearest')
-    ax_sofc_im.set_xlabel(f"Unknown material's phase fraction: {sofc_large_im.mean():.3f}")
+    ax_sofc_im.set_xlabel(f"Material's unknown phase fraction: {sofc_large_im.mean():.3f}", fontsize=11)
 
     patch_positions = []
     phase_fractions = []
@@ -74,7 +74,7 @@
     # Create the inset:
     inset_shift = 1 + wspace
     ax_inset = ax_sofc_im.inset_axes([inset_shift, 0, 1, 1], xlim=(x1, x2), ylim=(y1, y2))
-    ax_inset.set_xlabel(f"Sample's phase fraction: {sofc_small_im.mean():.3f}")
+    ax_inset.set_xlabel(f"Sample's phase fraction: {sofc_small_im.mean():.3f}", fontsize=11)
     inset_pos = ax_inset.get_position()
     ax_inset.imshow(sofc_small_im, cmap='gray', interpolation='nearest', extent=[x1, x2, y1, y2])
     for spine in ax_inset.spines.values():
@@ -86,8 +86,8 @@
     ax_sofc_im.set_yticks([])
     ax_inset.set_xticks([])
     ax_inset.set_yticks([])
-    ax_sofc_im.set_title("(a)")
-    ax_inset.set_title("(b)")
+    ax_sofc_im.set_title("(a)", fontsize=12)
+    ax_inset.set_title("(b)", fontsize=12)
 
 
 
@@ -99,37 +99,38 @@
     x = np.linspace(sofc_large_im.mean() - gap_pfs, sofc_large_im.mean() + gap_pfs, 200)
     p = norm.pdf(x, mu, sigma)
     unknown_dist.plot(x, p, color=COLOR_IN, linewidth=2, 
-                      label="Unknown phase fraction\ndistribution of same-sized samples.")
+                      label="Same-sized samples unknown\nphase fraction distribution.")
     # Make the plot 1.5 times lower in the plot:
-    unknown_dist.set_ylim(0, unknown_dist.get_ylim()[1]*1.6)
+    unknown_dist.set_ylim(0, unknown_dist.get_ylim()[1]*1.8)
     unknown_dist.set_xlim(sofc_large_im.mean() - gap_pfs, sofc_large_im.mean() + gap_pfs)
     unknown_dist.set_yticks([])
 
     # Now add the unknown material's phase fraction, until the normal distribution:
     ymax = norm.pdf(sofc_large_im.mean(), mu, sigma)
     ymax_mean = ymax / unknown_dist.get_ylim()[1]
-    unknown_dist.axvline(sofc_large_im.mean(), ymax=ymax_mean, color='black', linestyle='--', linewidth=2,
-                         label="Unknown material's phase fraction")
     # And the sample phase fraction:
     ymax_sample = norm.pdf(sofc_small_im.mean(), mu, sigma)
-    ymax_sample = ymax_sample / unknown_dist.get_ylim()[1]
+    ymax_sample = ymax_sample / unknown_dist.get_ylim()[1] 
     unknown_dist.axvline(sofc_small_im.mean(), ymax=ymax_sample, color=COLOR_INSET, linestyle='--', linewidth=2, 
                          label="Sample's phase fraction")
-    unknown_dist.legend(loc='upper left')
-    unknown_dist.set_title("(c)")
-    unknown_dist.set_xlabel('Phase fraction')
+    unknown_dist.axvline(sofc_large_im.mean(), ymax=ymax_mean, color='black', linestyle='--', linewidth=2,
+                         label="Material's unknown phase\nfraction")
+
+    unknown_dist.legend(loc='upper left', fontsize=11)
+    unknown_dist.set_title("(c)", fontsize=12)
+    unknown_dist.set_xlabel('Phase fraction', fontsize=11)
 
 
     # Rep. calculation:
     same_small_im = fig.add_subplot(gs[1, 0])
     same_small_im.imshow(sofc_small_im, cmap='gray', interpolation='nearest')
     same_small_im.set_xticks([])
     same_small_im.set_yticks([])
-    same_small_im.set_xlabel(f"Single image input")
+    same_small_im.set_xlabel(f"Single image input", fontsize=11)
     for spine in same_small_im.spines.values():
         spine.set_edgecolor(COLOR_INSET)
         spine.set_linewidth(LINE_W)
-    same_small_im.set_title('(e)')
+    same_small_im.set_title('(e)', fontsize=12)
 
     # Create the TPC plot:
     tpc_plot = fig.add_subplot(gs[1, 1])
@@ -138,12 +139,12 @@
     tpc_plot.set_ylabel('')
     tpc_plot.set_xticks([])
     tpc_plot.set_yticks([])
-    tpc_plot.set_title("(f)")
+    tpc_plot.set_title("(f)", fontsize=12)
 
     # Text of representativity analysis:
     rep_text = fig.add_subplot(gs[1, 2])
 
-    microlib_examples = tifffile.imread("paper_figures/microlib_examples.tif")
+    microlib_examples = tifffile.imread("paper_figures/figure_data/microlib_examples.tif")
     microlib_examples_size = min(microlib_examples.shape)
     microlib_examples = microlib_examples[-microlib_examples_size:, -microlib_examples_size:]
     microlib_examples = microlib_examples / microlib_examples.max()
@@ -153,13 +154,13 @@
     large_microlib_im[-microlib_examples_size:, start_idx:start_idx+microlib_examples_size] = microlib_examples
     rep_text.imshow(large_microlib_im, cmap='gray', interpolation='nearest')
 
-    rep_text.set_title("(g)")
+    rep_text.set_title("(g)", fontsize=12)
     text_y = 0.1
     text_pos = (0.5*large_microlib_im.shape[1], text_y*large_microlib_im.shape[0])
     rep_text.text(*text_pos, 
                   "TPC integration and data-driven \ncorrection using MicroLib ", va='center', ha='center')
     font_size = rep_text.texts[0].get_fontsize()
-    rep_text.texts[0].set_fontsize(font_size + 2)
+    rep_text.texts[0].set_fontsize(font_size + 3)
     # delete the axis:
     rep_text.axis('off')
 
@@ -174,10 +175,10 @@
     conf_bounds, pf_1d, cum_sum_sum = get_prediction_interval_stats(sofc_small_im)
     # cumulative sum to the original data:
     original_data = np.diff(cum_sum_sum)
-    res_1_pred.plot(pf_1d[1:], original_data, label="Predicted likelihood of material's\nphase fraction", linewidth=2)
-    res_1_pred.set_ylim(0, res_1_pred.get_ylim()[1]*1.6)
+    res_1_pred.plot(pf_1d[1:], original_data, label="Predicted likelihood of\nmaterial's phase fraction", linewidth=2)
+    res_1_pred.set_ylim(0, res_1_pred.get_ylim()[1]*1.8)
     res_1_pred.set_xlim(pf_1d[0], pf_1d[-1])
-    res_1_pred.set_xlabel('Phase fraction')
+    res_1_pred.set_xlabel('Phase fraction', fontsize=11)
     # No y-ticks:
     # res_1_pred.set_yticks([])
     # Fill between confidence bounds:
@@ -212,19 +213,19 @@
         linestyle="--",
         linewidth=2,
         color='black',
-        label="Unknown material's phase fraction",
+        label="Material's unknown phase\nfraction",
     )
     # No y-ticks:
     res_1_pred.set_yticks([])
-    res_1_pred.set_title("(d)")
+    res_1_pred.set_title("(d)", fontsize=12)
 
     res_1_pred.set_ylim([0, res_1_pred.get_ylim()[1]])
     inset_pf = sofc_small_im.mean()
     xerr = inset_pf - conf_start
     res_1_pred.errorbar(
         sofc_small_im.mean(), 0.0003, xerr=xerr, fmt='o', capsize=6, 
         color=COLOR_INSET, label="95% confidence interval", linewidth=LINE_W, capthick=LINE_W)
-    res_1_pred.legend(loc='upper left')
+    res_1_pred.legend(loc='upper left', fontsize=11)
 
 
 
@@ -267,7 +268,7 @@
     res_2_pred.set_xticks([])
     res_2_pred.set_yticks([])
 
-    label = f"Predicted sample size for a smaller\n95% confidence interval, with at\nmost {target_percetange_error}% deviation from material's\nphase fraction instead of {np.round(percent_error*100, 2)}%."
+    label = f"Sample size needed for ±{target_percetange_error}%\ndeviation from Material's phase\nfraction (95% CI), vs. current\n±{np.round(percent_error*100, 2)}% deviation (± 0.022)."
     # Create custom Line2D objects for the spines
     spine_line = mlines.Line2D([], [], color='black', linestyle='--', 
                                linewidth=2, alpha=alpha, 
@@ -287,14 +288,14 @@
                                 linewidth=1)
 
     # Add the custom lines to the legend
-    res_2_pred.legend(handles=[spine_line], loc='upper left')
+    res_2_pred.legend(handles=[spine_line], loc='upper left', fontsize=11)
 
     for spine in res_2_pred.spines.values():
         spine.set_linestyle("--")
         spine.set_linewidth(2)
         spine.set_alpha(alpha)
 
-    res_2_pred.set_title("(h)")
+    res_2_pred.set_title("(h)", fontsize=12)
 
     positions = []
     for i in range(8):
@@ -356,7 +357,7 @@
         # Insert the title as text in the top middle of the rectangle:
         text_pos = (lower_left_corner[0] + width_rect/2, lower_left_corner[1] + height_rect - gap_up_down/5)
         fig.text(text_pos[0], text_pos[1], title, ha='center', va='center', 
-                 fontsize=font_size+4, transform=fig.transFigure)
+                 fontsize=18, transform=fig.transFigure)
 
         fig.patches.append(rect)
 

paper_figures/model_accuracy.py

@@ -14,7 +14,7 @@
 LINE_W = 1.5
 
 
-def get_prediction_interval_stats(inset_image):
+def get_prediction_interval_stats(inset_image, conf=0.95):
     phase_fraction = float(np.mean(inset_image))
     n_dims = len(inset_image.shape)  # 2D or 3D
     n_elems = int(np.prod(inset_image.shape))
@@ -42,6 +42,7 @@ def get_prediction_interval_stats(inset_image):
         inset_image.mean(),
         std_bern,
         std_model,
+        conf_level=conf,
     )
 
     return conf_bounds, pf_1d, cum_sum_sum