Spaces:
Sleeping
Sleeping
rmm
commited on
Commit
·
4854d2c
1
Parent(s):
7a5f0ca
feat: using FSM for full workflow, with some steps mocked
Browse files- dropped the "ML running" phase for now as we don't do it async
- src/classifier/classifier_image.py +9 -0
- src/main.py +123 -37
- src/utils/workflow_state.py +17 -1
src/classifier/classifier_image.py
CHANGED
@@ -11,6 +11,15 @@ from hf_push_observations import push_observations
|
|
11 |
from utils.grid_maker import gridder
|
12 |
from utils.metadata_handler import metadata2md
|
13 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
def cetacean_classify(cetacean_classifier):
|
15 |
"""Cetacean classifier using the saving-willy model from Saving Willy Hugging Face space.
|
16 |
For each image in the session state, classify the image and display the top 3 predictions.
|
|
|
11 |
from utils.grid_maker import gridder
|
12 |
from utils.metadata_handler import metadata2md
|
13 |
|
14 |
+
def add_header_text() -> None:
|
15 |
+
"""
|
16 |
+
Add brief explainer text about cetacean classification to the tab
|
17 |
+
"""
|
18 |
+
st.markdown("""
|
19 |
+
*Run classifer to identify the species of cetean on the uploaded image.
|
20 |
+
Once inference is complete, the top three predictions are shown.
|
21 |
+
You can override the prediction by selecting a species from the dropdown.*""")
|
22 |
+
|
23 |
def cetacean_classify(cetacean_classifier):
|
24 |
"""Cetacean classifier using the saving-willy model from Saving Willy Hugging Face space.
|
25 |
For each image in the session state, classify the image and display the top 3 predictions.
|
src/main.py
CHANGED
@@ -9,7 +9,8 @@ from streamlit_folium import st_folium
|
|
9 |
from transformers import pipeline
|
10 |
from transformers import AutoModelForImageClassification
|
11 |
|
12 |
-
from maps.obs_map import add_header_text
|
|
|
13 |
from datasets import disable_caching
|
14 |
disable_caching()
|
15 |
|
@@ -79,18 +80,20 @@ if "workflow_fsm" not in st.session_state:
|
|
79 |
# create and init the state machine
|
80 |
st.session_state.workflow_fsm = WorkflowFSM(FSM_STATES)
|
81 |
|
82 |
-
# add progress indicator to session_state
|
83 |
-
if "progress" not in st.session_state:
|
84 |
-
with st.sidebar:
|
85 |
-
st.session_state.disp_progress = [st.empty(), st.empty()]
|
86 |
-
|
87 |
def refresh_progress():
|
88 |
with st.sidebar:
|
89 |
-
tot = st.session_state.workflow_fsm.num_states
|
90 |
cur_i = st.session_state.workflow_fsm.current_state_index
|
91 |
cur_t = st.session_state.workflow_fsm.current_state
|
92 |
st.session_state.disp_progress[0].markdown(f"*Progress: {cur_i}/{tot}. Current: {cur_t}.*")
|
93 |
st.session_state.disp_progress[1].progress(cur_i/tot)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
94 |
|
95 |
|
96 |
def main() -> None:
|
@@ -125,10 +128,8 @@ def main() -> None:
|
|
125 |
st.tabs(["Cetecean classifier", "Hotdog classifier", "Map", "*:gray[Dev:coordinates]*", "Log", "Beautiful cetaceans"])
|
126 |
st.session_state.tab_log = tab_log
|
127 |
|
|
|
128 |
refresh_progress()
|
129 |
-
# add button to sidebar, with the callback to refesh_progress
|
130 |
-
st.sidebar.button("Refresh Progress", on_click=refresh_progress)
|
131 |
-
|
132 |
|
133 |
# create a sidebar, and parse all the input (returned as `observations` object)
|
134 |
setup_input(viewcontainer=st.sidebar)
|
@@ -149,7 +150,7 @@ def main() -> None:
|
|
149 |
with tab_map:
|
150 |
# visual structure: a couple of toggles at the top, then the map inlcuding a
|
151 |
# dropdown for tileset selection.
|
152 |
-
|
153 |
tab_map_ui_cols = st.columns(2)
|
154 |
with tab_map_ui_cols[0]:
|
155 |
show_db_points = st.toggle("Show Points from DB", True)
|
@@ -207,24 +208,108 @@ def main() -> None:
|
|
207 |
gallery.render_whale_gallery(n_cols=4)
|
208 |
|
209 |
|
210 |
-
#
|
211 |
-
|
212 |
-
|
213 |
-
|
214 |
-
|
215 |
-
|
216 |
-
|
|
|
|
|
|
|
|
|
|
|
217 |
st.session_state.workflow_fsm.complete_current_state()
|
218 |
-
|
219 |
-
|
220 |
-
|
221 |
-
|
222 |
-
|
|
|
|
|
|
|
|
|
223 |
# create a dictionary with the submitted observation
|
224 |
tab_log.info(f"{st.session_state.observations}")
|
225 |
df = pd.DataFrame(st.session_state.observations, index=[0])
|
226 |
with tab_coords:
|
227 |
st.table(df)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
228 |
|
229 |
|
230 |
|
@@ -235,23 +320,24 @@ def main() -> None:
|
|
235 |
# - these species are shown
|
236 |
# - the user can override the species prediction using the dropdown
|
237 |
# - an observation is uploaded if the user chooses.
|
238 |
-
|
239 |
-
|
240 |
-
|
241 |
-
You can override the prediction by selecting a species from the dropdown.*""")
|
242 |
|
243 |
-
if tab_inference.button("Identify with cetacean classifier"):
|
244 |
-
#pipe = pipeline("image-classification", model="Saving-Willy/cetacean-classifier", trust_remote_code=True)
|
245 |
-
cetacean_classifier = AutoModelForImageClassification.from_pretrained("Saving-Willy/cetacean-classifier",
|
246 |
-
revision=classifier_revision,
|
247 |
-
trust_remote_code=True)
|
248 |
|
249 |
|
250 |
-
|
251 |
-
|
252 |
-
|
253 |
-
|
254 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
255 |
|
256 |
|
257 |
|
|
|
9 |
from transformers import pipeline
|
10 |
from transformers import AutoModelForImageClassification
|
11 |
|
12 |
+
from maps.obs_map import add_header_text as add_obs_map_header
|
13 |
+
from classifier.classifier_image import add_header_text as add_classifier_header
|
14 |
from datasets import disable_caching
|
15 |
disable_caching()
|
16 |
|
|
|
80 |
# create and init the state machine
|
81 |
st.session_state.workflow_fsm = WorkflowFSM(FSM_STATES)
|
82 |
|
|
|
|
|
|
|
|
|
|
|
83 |
def refresh_progress():
|
84 |
with st.sidebar:
|
85 |
+
tot = st.session_state.workflow_fsm.num_states - 1
|
86 |
cur_i = st.session_state.workflow_fsm.current_state_index
|
87 |
cur_t = st.session_state.workflow_fsm.current_state
|
88 |
st.session_state.disp_progress[0].markdown(f"*Progress: {cur_i}/{tot}. Current: {cur_t}.*")
|
89 |
st.session_state.disp_progress[1].progress(cur_i/tot)
|
90 |
+
# add progress indicator to session_state
|
91 |
+
if "progress" not in st.session_state:
|
92 |
+
with st.sidebar:
|
93 |
+
st.session_state.disp_progress = [st.empty(), st.empty()]
|
94 |
+
# add button to sidebar, with the callback to refesh_progress
|
95 |
+
st.sidebar.button("Refresh Progress", on_click=refresh_progress)
|
96 |
+
|
97 |
|
98 |
|
99 |
def main() -> None:
|
|
|
128 |
st.tabs(["Cetecean classifier", "Hotdog classifier", "Map", "*:gray[Dev:coordinates]*", "Log", "Beautiful cetaceans"])
|
129 |
st.session_state.tab_log = tab_log
|
130 |
|
131 |
+
# put this early so the progress indicator is at the top (also refreshed at end)
|
132 |
refresh_progress()
|
|
|
|
|
|
|
133 |
|
134 |
# create a sidebar, and parse all the input (returned as `observations` object)
|
135 |
setup_input(viewcontainer=st.sidebar)
|
|
|
150 |
with tab_map:
|
151 |
# visual structure: a couple of toggles at the top, then the map inlcuding a
|
152 |
# dropdown for tileset selection.
|
153 |
+
add_obs_map_header()
|
154 |
tab_map_ui_cols = st.columns(2)
|
155 |
with tab_map_ui_cols[0]:
|
156 |
show_db_points = st.toggle("Show Points from DB", True)
|
|
|
208 |
gallery.render_whale_gallery(n_cols=4)
|
209 |
|
210 |
|
211 |
+
# state handling re data_entry phases
|
212 |
+
# 0. no data entered yet -> display the file uploader thing
|
213 |
+
# 1. we have some images, but not all the metadata fields are done -> validate button shown, disabled
|
214 |
+
# 2. all data entered -> validate button enabled
|
215 |
+
# 3. validation button pressed, validation done -> enable the inference button.
|
216 |
+
# - at this point do we also want to disable changes to the metadata selectors?
|
217 |
+
# anyway, simple first.
|
218 |
+
|
219 |
+
if st.session_state.workflow_fsm.is_in_state('doing_data_entry'):
|
220 |
+
# can we advance state? - only when all inputs are set for all uploaded files
|
221 |
+
all_inputs_set = check_inputs_are_set(debug=True)
|
222 |
+
if all_inputs_set:
|
223 |
st.session_state.workflow_fsm.complete_current_state()
|
224 |
+
# -> data_entry_complete
|
225 |
+
else:
|
226 |
+
# button, disabled; no state change yet.
|
227 |
+
st.sidebar.button(":gray[*Validate*]", disabled=True, help="Please fill in all fields.")
|
228 |
+
|
229 |
+
|
230 |
+
if st.session_state.workflow_fsm.is_in_state('data_entry_complete'):
|
231 |
+
# can we advance state? - only when the validate button is pressed
|
232 |
+
if st.sidebar.button(":white_check_mark:[*Validate*]"):
|
233 |
# create a dictionary with the submitted observation
|
234 |
tab_log.info(f"{st.session_state.observations}")
|
235 |
df = pd.DataFrame(st.session_state.observations, index=[0])
|
236 |
with tab_coords:
|
237 |
st.table(df)
|
238 |
+
# there doesn't seem to be any actual validation here?? TODO: find validator function (each element is validated by the input box, but is there something at the whole image level?)
|
239 |
+
# hmm, maybe it should actually just be "I'm done with data entry"
|
240 |
+
st.session_state.workflow_fsm.complete_current_state()
|
241 |
+
# -> data_entry_validated
|
242 |
+
|
243 |
+
# state handling re inference phases (tab_inference)
|
244 |
+
# 3. validation button pressed, validation done -> enable the inference button.
|
245 |
+
# 4. inference button pressed -> ML started. | let's cut this one out, since it would only
|
246 |
+
# make sense if we did it as an async action
|
247 |
+
# 5. ML done -> show results, and manual validation options
|
248 |
+
# 6. manual validation done -> enable the upload buttons
|
249 |
+
#
|
250 |
+
with tab_inference:
|
251 |
+
add_classifier_header()
|
252 |
+
# if we are before data_entry_validated, show the button, disabled.
|
253 |
+
if not st.session_state.workflow_fsm.is_in_state_or_beyond('data_entry_validated'):
|
254 |
+
tab_inference.button(":gray[*Identify with cetacean classifier*]", disabled=True,
|
255 |
+
help="Please validate inputs before proceeding",
|
256 |
+
key="button_infer_ceteans")
|
257 |
+
|
258 |
+
if st.session_state.workflow_fsm.is_in_state('data_entry_validated'):
|
259 |
+
# show the button, enabled. If pressed, we start the ML model (And advance state)
|
260 |
+
if tab_inference.button("Identify with cetacean classifier"):
|
261 |
+
cetacean_classifier = AutoModelForImageClassification.from_pretrained(
|
262 |
+
"Saving-Willy/cetacean-classifier",
|
263 |
+
revision=classifier_revision,
|
264 |
+
trust_remote_code=True)
|
265 |
+
|
266 |
+
cetacean_classify(cetacean_classifier)
|
267 |
+
st.session_state.workflow_fsm.complete_current_state()
|
268 |
+
|
269 |
+
if st.session_state.workflow_fsm.is_in_state('ml_classification_completed'):
|
270 |
+
# show the results, and allow manual validation
|
271 |
+
s = ""
|
272 |
+
for k, v in st.session_state.whale_prediction1.items():
|
273 |
+
s += f"* Image {k}: {v}\n"
|
274 |
+
|
275 |
+
st.markdown("""
|
276 |
+
### Inference Results and manual validation/adjustment
|
277 |
+
:construction: for now we just show the num images processed.
|
278 |
+
""")
|
279 |
+
st.markdown(s)
|
280 |
+
# add a button to advance the state
|
281 |
+
if st.button("mock: manual validation done."):
|
282 |
+
st.session_state.workflow_fsm.complete_current_state()
|
283 |
+
# -> manual_inspection_completed
|
284 |
+
|
285 |
+
if st.session_state.workflow_fsm.is_in_state('manual_inspection_completed'):
|
286 |
+
# show the ML results, and allow the user to upload the observation
|
287 |
+
st.markdown("""
|
288 |
+
### Inference Results (after manual validation)
|
289 |
+
:construction: for now we just show the button.
|
290 |
+
""")
|
291 |
+
|
292 |
+
|
293 |
+
if st.button("(nooop) Upload observation to THE INTERNET!"):
|
294 |
+
st.session_state.workflow_fsm.complete_current_state()
|
295 |
+
# -> data_uploaded
|
296 |
+
|
297 |
+
if st.session_state.workflow_fsm.is_in_state('data_uploaded'):
|
298 |
+
# the data has been sent. Lets show the observations again
|
299 |
+
# but no buttons to upload (or greyed out ok)
|
300 |
+
st.markdown("""
|
301 |
+
### Observation(s) uploaded
|
302 |
+
:construction: for now we just show the observations.
|
303 |
+
""")
|
304 |
+
df = pd.DataFrame(st.session_state.observations, index=[0])
|
305 |
+
st.table(df)
|
306 |
+
|
307 |
+
# didn't decide what the next state is here - I think we are in the terminal state.
|
308 |
+
#st.session_state.workflow_fsm.complete_current_state()
|
309 |
+
|
310 |
+
|
311 |
+
|
312 |
+
|
313 |
|
314 |
|
315 |
|
|
|
320 |
# - these species are shown
|
321 |
# - the user can override the species prediction using the dropdown
|
322 |
# - an observation is uploaded if the user chooses.
|
323 |
+
|
324 |
+
# with tab_inference:
|
325 |
+
# add_classifier_header()
|
|
|
326 |
|
|
|
|
|
|
|
|
|
|
|
327 |
|
328 |
|
329 |
+
# if tab_inference.button("Identify with cetacean classifier"):
|
330 |
+
# #pipe = pipeline("image-classification", model="Saving-Willy/cetacean-classifier", trust_remote_code=True)
|
331 |
+
# cetacean_classifier = AutoModelForImageClassification.from_pretrained("Saving-Willy/cetacean-classifier",
|
332 |
+
# revision=classifier_revision,
|
333 |
+
# trust_remote_code=True)
|
334 |
+
|
335 |
+
|
336 |
+
# if st.session_state.images is None:
|
337 |
+
# # TODO: cleaner design to disable the button until data input done?
|
338 |
+
# st.info("Please upload an image first.")
|
339 |
+
# else:
|
340 |
+
# cetacean_classify(cetacean_classifier)
|
341 |
|
342 |
|
343 |
|
src/utils/workflow_state.py
CHANGED
@@ -8,7 +8,10 @@ FAIL = '\033[91m'
|
|
8 |
ENDC = '\033[0m'
|
9 |
|
10 |
|
11 |
-
FSM_STATES = ['
|
|
|
|
|
|
|
12 |
|
13 |
|
14 |
class WorkflowFSM:
|
@@ -64,6 +67,19 @@ class WorkflowFSM:
|
|
64 |
return False
|
65 |
return False
|
66 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
67 |
@property
|
68 |
def current_state(self) -> str:
|
69 |
"""Get the current state name"""
|
|
|
8 |
ENDC = '\033[0m'
|
9 |
|
10 |
|
11 |
+
FSM_STATES = ['doing_data_entry', 'data_entry_complete', 'data_entry_validated',
|
12 |
+
#'ml_classification_started',
|
13 |
+
'ml_classification_completed',
|
14 |
+
'manual_inspection_completed', 'data_uploaded']
|
15 |
|
16 |
|
17 |
class WorkflowFSM:
|
|
|
67 |
return False
|
68 |
return False
|
69 |
|
70 |
+
# add a helper method, to find out if a given state has been reached/passed
|
71 |
+
# we first need to get the index of the current state
|
72 |
+
# then the index of the argument state
|
73 |
+
# compare, and return boolean
|
74 |
+
|
75 |
+
def is_in_state_or_beyond(self, state_name: str) -> bool:
|
76 |
+
"""Check if we have reached or passed the specified state"""
|
77 |
+
if state_name not in self.state_dict:
|
78 |
+
raise ValueError(f"Invalid state: {state_name}")
|
79 |
+
|
80 |
+
return self.state_dict[state_name] <= self.state_dict[self.state]
|
81 |
+
|
82 |
+
|
83 |
@property
|
84 |
def current_state(self) -> str:
|
85 |
"""Get the current state name"""
|