visualize_page method in demo/app.py file

[khawar-islam]

I have a question on visualize_page method, it takes two parameters to visualize full output in an app. Are there any possibility to display only specific output. I have extracted some special information from image i do not want to display all the information extracted by OCR.

Code

out = predictor([doc[page_idx]])
                export = out.export()
                page_words = [[word for block in page['blocks'] for line in block['lines'] for word in line['words']]
                              for page
                              in export['pages']]

                page_dims = [page['dimensions'] for page in export['pages']]

                for words, dims in zip(page_words, page_dims):
                    for word in page_words:
                        for single_word in word:
                            if '-' in single_word['value']:
                                xmin = int(single_word['geometry'][0][0] * dims[1])
                                ymin = int(single_word['geometry'][0][1] * dims[0])
                                xmax = int(single_word['geometry'][1][0] * dims[1])
                                ymax = int(single_word['geometry'][1][1] * dims[0])
                                image = cv2.rectangle(doc[page_idx], (xmin, ymin), (xmax, ymax), (224, 224, 153),
                                                      cv2.FILLED)
                                #cv2.imwrite(os.path.join(result_path, file_Name), img_bgr_rgb)
                            else:
                                print(single_word)
                                #cv2.imwrite(os.path.join(result_path, file_Name), img_bgr_rgb)

                print(out.pages[0].export())
                fig = visualize_page(out.pages[0].export(), image, interactive=False, add_labels=False)

                cols[1].pyplot(fig)

Your visualize_page method is taking all out.pages[0].export() in first parameter. Can we ignore that one? I have tried multiple time but failed.

fig = visualize_page(out.pages[0].export(), image, interactive=False, add_labels=False)

Output
I do not need the highlighted blue part.

[felixdittrich92]

Hi @khawar-islam 👋,
have you tried to .pop() the entries you don't want to visualize ?

[khawar-islam]

@felixdittrich92 I did not tried .pop() but i tried something different and it works. I simply pass the image instance to cols[1].image(image). Still .json is incorrect form.

            with st.spinner('Analyzing...'):

                # Plot CMAX output
                out = predictor([doc[page_idx]])

                export = out.export()

                page_words = [[word for block in page['blocks'] for line in block['lines'] for word in line['words']]
                              for page
                              in export['pages']]

                page_dims = [page['dimensions'] for page in export['pages']]

                for words, dims in zip(page_words, page_dims):
                    for word in page_words:
                        for single_word in word:
                            if '-' in single_word['value']:
                                print(single_word['value'])
                                xmin = int(single_word['geometry'][0][0] * dims[1])
                                ymin = int(single_word['geometry'][0][1] * dims[0])
                                xmax = int(single_word['geometry'][1][0] * dims[1])
                                ymax = int(single_word['geometry'][1][1] * dims[0])
                                image = cv2.rectangle(doc[page_idx], (xmin, ymin), (xmax, ymax), (173, 216, 230), cv2.FILLED)
                                plt.imsave(os.path.abspath(os.getcwd()) + "/coremax_img/" + uploaded_file.name, image)
                            else:
                                plt.imsave(os.path.abspath(os.getcwd()) + "/coremax_img/" + uploaded_file.name, doc[page_idx])

                cols[1].image(image)

[felixdittrich92]

Nice that you was able to solve it :)

[frgfm]

Also, to try to better understand, in add on Felix's suggestion, I have a question @khawar-islam
You did explain what you don't want, but I'm honestly not certain about what you expect then?

• the JSON?
• the unchanged image?
• the synthesized image?

[felixdittrich92]

If i understood it correctly than he want to display only the masked parts @khawar-islam could you confirm this ? 😅

[khawar-islam]

Yes, only masked part those rectangle boxes which are filled.

[frgfm]

So you want the list of images that corresponds to those crops? 🤔